I can use text to get audio response and barge in

.env.example

@@ -0,0 +1,30 @@
+# Server Configuration
+HOST=0.0.0.0
+PORT=8000
+
+# Audio Configuration
+SAMPLE_RATE=16000
+CHUNK_SIZE_MS=20
+
+# VAD Configuration
+VAD_THRESHOLD=0.5
+VAD_EOU_THRESHOLD_MS=400
+
+# OpenAI / LLM Configuration (required for duplex voice)
+OPENAI_API_KEY=sk-your-openai-api-key-here
+# OPENAI_API_URL=https://api.openai.com/v1  # Optional: for Azure or compatible APIs
+LLM_MODEL=gpt-4o-mini
+LLM_TEMPERATURE=0.7
+
+# TTS Configuration
+TTS_VOICE=en-US-JennyNeural
+TTS_SPEED=1.0
+
+# Duplex Pipeline Configuration
+DUPLEX_ENABLED=true
+# DUPLEX_GREETING=Hello! How can I help you today?
+DUPLEX_SYSTEM_PROMPT=You are a helpful, friendly voice assistant. Keep your responses concise and conversational.
+
+# Logging
+LOG_LEVEL=INFO
+LOG_FORMAT=text

app/config.py

@@ -33,6 +33,29 @@ class Settings(BaseSettings):
     vad_min_speech_duration_ms: int = Field(default=250, description="Minimum speech duration in milliseconds")
     vad_eou_threshold_ms: int = Field(default=400, description="End of utterance (silence) threshold in milliseconds")
 
+    # OpenAI / LLM Configuration
+    openai_api_key: Optional[str] = Field(default=None, description="OpenAI API key")
+    openai_api_url: Optional[str] = Field(default=None, description="OpenAI API base URL (for Azure/compatible)")
+    llm_model: str = Field(default="gpt-4o-mini", description="LLM model name")
+    llm_temperature: float = Field(default=0.7, description="LLM temperature for response generation")
+    
+    # TTS Configuration
+    tts_provider: str = Field(default="siliconflow", description="TTS provider (edge, siliconflow)")
+    tts_voice: str = Field(default="anna", description="TTS voice name")
+    tts_speed: float = Field(default=1.0, description="TTS speech speed multiplier")
+    
+    # SiliconFlow Configuration
+    siliconflow_api_key: Optional[str] = Field(default=None, description="SiliconFlow API key")
+    siliconflow_tts_model: str = Field(default="FunAudioLLM/CosyVoice2-0.5B", description="SiliconFlow TTS model")
+    
+    # Duplex Pipeline Configuration
+    duplex_enabled: bool = Field(default=True, description="Enable duplex voice pipeline")
+    duplex_greeting: Optional[str] = Field(default=None, description="Optional greeting message")
+    duplex_system_prompt: Optional[str] = Field(
+        default="You are a helpful, friendly voice assistant. Keep your responses concise and conversational.",
+        description="System prompt for LLM"
+    )
+
     # Logging
     log_level: str = Field(default="INFO", description="Logging level")
     log_format: str = Field(default="json", description="Log format (json or text)")

core/__init__.py

@@ -1 +1,22 @@
 """Core Components Package"""
+
+from core.events import EventBus, get_event_bus
+from core.transports import BaseTransport, SocketTransport, WebRtcTransport
+from core.pipeline import AudioPipeline
+from core.session import Session
+from core.conversation import ConversationManager, ConversationState, ConversationTurn
+from core.duplex_pipeline import DuplexPipeline
+
+__all__ = [
+    "EventBus",
+    "get_event_bus",
+    "BaseTransport",
+    "SocketTransport",
+    "WebRtcTransport",
+    "AudioPipeline",
+    "Session",
+    "ConversationManager",
+    "ConversationState",
+    "ConversationTurn",
+    "DuplexPipeline",
+]

core/conversation.py

@@ -0,0 +1,255 @@
+"""Conversation management for voice AI.
+
+Handles conversation context, turn-taking, and message history
+for multi-turn voice conversations.
+"""
+
+import asyncio
+from typing import List, Optional, Dict, Any, Callable, Awaitable
+from dataclasses import dataclass, field
+from enum import Enum
+from loguru import logger
+
+from services.base import LLMMessage
+
+
+class ConversationState(Enum):
+    """State of the conversation."""
+    IDLE = "idle"                    # Waiting for user input
+    LISTENING = "listening"          # User is speaking
+    PROCESSING = "processing"        # Processing user input (LLM)
+    SPEAKING = "speaking"            # Bot is speaking
+    INTERRUPTED = "interrupted"      # Bot was interrupted
+
+
+@dataclass
+class ConversationTurn:
+    """A single turn in the conversation."""
+    role: str  # "user" or "assistant"
+    text: str
+    audio_duration_ms: Optional[int] = None
+    timestamp: float = field(default_factory=lambda: asyncio.get_event_loop().time())
+    was_interrupted: bool = False
+
+
+class ConversationManager:
+    """
+    Manages conversation state and history.
+    
+    Provides:
+    - Message history for LLM context
+    - Turn management
+    - State tracking
+    - Event callbacks for state changes
+    """
+    
+    def __init__(
+        self,
+        system_prompt: Optional[str] = None,
+        max_history: int = 20,
+        greeting: Optional[str] = None
+    ):
+        """
+        Initialize conversation manager.
+        
+        Args:
+            system_prompt: System prompt for LLM
+            max_history: Maximum number of turns to keep
+            greeting: Optional greeting message when conversation starts
+        """
+        self.system_prompt = system_prompt or (
+            "You are a helpful, friendly voice assistant. "
+            "Keep your responses concise and conversational. "
+            "Respond naturally as if having a phone conversation. "
+            "If you don't understand something, ask for clarification."
+        )
+        self.max_history = max_history
+        self.greeting = greeting
+        
+        # State
+        self.state = ConversationState.IDLE
+        self.turns: List[ConversationTurn] = []
+        
+        # Callbacks
+        self._state_callbacks: List[Callable[[ConversationState, ConversationState], Awaitable[None]]] = []
+        self._turn_callbacks: List[Callable[[ConversationTurn], Awaitable[None]]] = []
+        
+        # Current turn tracking
+        self._current_user_text: str = ""
+        self._current_assistant_text: str = ""
+        
+        logger.info("ConversationManager initialized")
+    
+    def on_state_change(
+        self,
+        callback: Callable[[ConversationState, ConversationState], Awaitable[None]]
+    ) -> None:
+        """Register callback for state changes."""
+        self._state_callbacks.append(callback)
+    
+    def on_turn_complete(
+        self,
+        callback: Callable[[ConversationTurn], Awaitable[None]]
+    ) -> None:
+        """Register callback for turn completion."""
+        self._turn_callbacks.append(callback)
+    
+    async def set_state(self, new_state: ConversationState) -> None:
+        """Set conversation state and notify listeners."""
+        if new_state != self.state:
+            old_state = self.state
+            self.state = new_state
+            logger.debug(f"Conversation state: {old_state.value} -> {new_state.value}")
+            
+            for callback in self._state_callbacks:
+                try:
+                    await callback(old_state, new_state)
+                except Exception as e:
+                    logger.error(f"State callback error: {e}")
+    
+    def get_messages(self) -> List[LLMMessage]:
+        """
+        Get conversation history as LLM messages.
+        
+        Returns:
+            List of LLMMessage objects including system prompt
+        """
+        messages = [LLMMessage(role="system", content=self.system_prompt)]
+        
+        # Add conversation history
+        for turn in self.turns[-self.max_history:]:
+            messages.append(LLMMessage(role=turn.role, content=turn.text))
+        
+        # Add current user text if any
+        if self._current_user_text:
+            messages.append(LLMMessage(role="user", content=self._current_user_text))
+        
+        return messages
+    
+    async def start_user_turn(self) -> None:
+        """Signal that user has started speaking."""
+        await self.set_state(ConversationState.LISTENING)
+        self._current_user_text = ""
+    
+    async def update_user_text(self, text: str, is_final: bool = False) -> None:
+        """
+        Update current user text (from ASR).
+        
+        Args:
+            text: Transcribed text
+            is_final: Whether this is the final transcript
+        """
+        self._current_user_text = text
+    
+    async def end_user_turn(self, text: str) -> None:
+        """
+        End user turn and add to history.
+        
+        Args:
+            text: Final user text
+        """
+        if text.strip():
+            turn = ConversationTurn(role="user", text=text.strip())
+            self.turns.append(turn)
+            
+            for callback in self._turn_callbacks:
+                try:
+                    await callback(turn)
+                except Exception as e:
+                    logger.error(f"Turn callback error: {e}")
+            
+            logger.info(f"User: {text[:50]}...")
+        
+        self._current_user_text = ""
+        await self.set_state(ConversationState.PROCESSING)
+    
+    async def start_assistant_turn(self) -> None:
+        """Signal that assistant has started speaking."""
+        await self.set_state(ConversationState.SPEAKING)
+        self._current_assistant_text = ""
+    
+    async def update_assistant_text(self, text: str) -> None:
+        """
+        Update current assistant text (streaming).
+        
+        Args:
+            text: Text chunk from LLM
+        """
+        self._current_assistant_text += text
+    
+    async def end_assistant_turn(self, was_interrupted: bool = False) -> None:
+        """
+        End assistant turn and add to history.
+        
+        Args:
+            was_interrupted: Whether the turn was interrupted by user
+        """
+        text = self._current_assistant_text.strip()
+        if text:
+            turn = ConversationTurn(
+                role="assistant",
+                text=text,
+                was_interrupted=was_interrupted
+            )
+            self.turns.append(turn)
+            
+            for callback in self._turn_callbacks:
+                try:
+                    await callback(turn)
+                except Exception as e:
+                    logger.error(f"Turn callback error: {e}")
+            
+            status = " (interrupted)" if was_interrupted else ""
+            logger.info(f"Assistant{status}: {text[:50]}...")
+        
+        self._current_assistant_text = ""
+        
+        if was_interrupted:
+            await self.set_state(ConversationState.INTERRUPTED)
+        else:
+            await self.set_state(ConversationState.IDLE)
+    
+    async def interrupt(self) -> None:
+        """Handle interruption (barge-in)."""
+        if self.state == ConversationState.SPEAKING:
+            await self.end_assistant_turn(was_interrupted=True)
+    
+    def reset(self) -> None:
+        """Reset conversation history."""
+        self.turns = []
+        self._current_user_text = ""
+        self._current_assistant_text = ""
+        self.state = ConversationState.IDLE
+        logger.info("Conversation reset")
+    
+    @property
+    def turn_count(self) -> int:
+        """Get number of turns in conversation."""
+        return len(self.turns)
+    
+    @property
+    def last_user_text(self) -> Optional[str]:
+        """Get last user text."""
+        for turn in reversed(self.turns):
+            if turn.role == "user":
+                return turn.text
+        return None
+    
+    @property
+    def last_assistant_text(self) -> Optional[str]:
+        """Get last assistant text."""
+        for turn in reversed(self.turns):
+            if turn.role == "assistant":
+                return turn.text
+        return None
+    
+    def get_context_summary(self) -> Dict[str, Any]:
+        """Get a summary of conversation context."""
+        return {
+            "state": self.state.value,
+            "turn_count": self.turn_count,
+            "last_user": self.last_user_text,
+            "last_assistant": self.last_assistant_text,
+            "current_user": self._current_user_text or None,
+            "current_assistant": self._current_assistant_text or None
+        }

core/duplex_pipeline.py

@@ -0,0 +1,509 @@
+"""Full duplex audio pipeline for AI voice conversation.
+
+This module implements the core duplex pipeline that orchestrates:
+- VAD (Voice Activity Detection)
+- EOU (End of Utterance) Detection
+- ASR (Automatic Speech Recognition) - optional
+- LLM (Language Model)
+- TTS (Text-to-Speech)
+
+Inspired by pipecat's frame-based architecture and active-call's
+event-driven design.
+"""
+
+import asyncio
+from typing import Optional, Callable, Awaitable
+from loguru import logger
+
+from core.transports import BaseTransport
+from core.conversation import ConversationManager, ConversationState
+from core.events import get_event_bus
+from processors.vad import VADProcessor, SileroVAD
+from processors.eou import EouDetector
+from services.base import BaseLLMService, BaseTTSService, BaseASRService
+from services.llm import OpenAILLMService, MockLLMService
+from services.tts import EdgeTTSService, MockTTSService
+from services.asr import BufferedASRService
+from services.siliconflow_tts import SiliconFlowTTSService
+from app.config import settings
+
+
+class DuplexPipeline:
+    """
+    Full duplex audio pipeline for AI voice conversation.
+    
+    Handles bidirectional audio flow with:
+    - User speech detection and transcription
+    - AI response generation
+    - Text-to-speech synthesis
+    - Barge-in (interruption) support
+    
+    Architecture (inspired by pipecat):
+    
+        User Audio → VAD → EOU → [ASR] → LLM → TTS → Audio Out
+                      ↓
+                  Barge-in Detection → Interrupt
+    """
+    
+    def __init__(
+        self,
+        transport: BaseTransport,
+        session_id: str,
+        llm_service: Optional[BaseLLMService] = None,
+        tts_service: Optional[BaseTTSService] = None,
+        asr_service: Optional[BaseASRService] = None,
+        system_prompt: Optional[str] = None,
+        greeting: Optional[str] = None
+    ):
+        """
+        Initialize duplex pipeline.
+        
+        Args:
+            transport: Transport for sending audio/events
+            session_id: Session identifier
+            llm_service: LLM service (defaults to OpenAI)
+            tts_service: TTS service (defaults to EdgeTTS)
+            asr_service: ASR service (optional)
+            system_prompt: System prompt for LLM
+            greeting: Optional greeting to speak on start
+        """
+        self.transport = transport
+        self.session_id = session_id
+        self.event_bus = get_event_bus()
+        
+        # Initialize VAD
+        self.vad_model = SileroVAD(
+            model_path=settings.vad_model_path,
+            sample_rate=settings.sample_rate
+        )
+        self.vad_processor = VADProcessor(
+            vad_model=self.vad_model,
+            threshold=settings.vad_threshold
+        )
+        
+        # Initialize EOU detector
+        self.eou_detector = EouDetector(
+            silence_threshold_ms=600,
+            min_speech_duration_ms=200
+        )
+        
+        # Initialize services
+        self.llm_service = llm_service
+        self.tts_service = tts_service
+        self.asr_service = asr_service or BufferedASRService()
+        
+        # Conversation manager
+        self.conversation = ConversationManager(
+            system_prompt=system_prompt,
+            greeting=greeting
+        )
+        
+        # State
+        self._running = True
+        self._is_bot_speaking = False
+        self._current_turn_task: Optional[asyncio.Task] = None
+        self._audio_buffer: bytes = b""
+        self._last_vad_status: str = "Silence"
+        
+        # Interruption handling
+        self._interrupt_event = asyncio.Event()
+        
+        logger.info(f"DuplexPipeline initialized for session {session_id}")
+    
+    async def start(self) -> None:
+        """Start the pipeline and connect services."""
+        try:
+            # Connect LLM service
+            if not self.llm_service:
+                if settings.openai_api_key:
+                    self.llm_service = OpenAILLMService(
+                        api_key=settings.openai_api_key,
+                        base_url=settings.openai_api_url,
+                        model=settings.llm_model
+                    )
+                else:
+                    logger.warning("No OpenAI API key - using mock LLM")
+                    self.llm_service = MockLLMService()
+            
+            await self.llm_service.connect()
+            
+            # Connect TTS service
+            if not self.tts_service:
+                if settings.tts_provider == "siliconflow" and settings.siliconflow_api_key:
+                    self.tts_service = SiliconFlowTTSService(
+                        api_key=settings.siliconflow_api_key,
+                        voice=settings.tts_voice,
+                        model=settings.siliconflow_tts_model,
+                        sample_rate=settings.sample_rate,
+                        speed=settings.tts_speed
+                    )
+                    logger.info("Using SiliconFlow TTS service")
+                else:
+                    self.tts_service = EdgeTTSService(
+                        voice=settings.tts_voice,
+                        sample_rate=settings.sample_rate
+                    )
+                    logger.info("Using Edge TTS service")
+            
+            await self.tts_service.connect()
+            
+            # Connect ASR service
+            await self.asr_service.connect()
+            
+            logger.info("DuplexPipeline services connected")
+            
+            # Speak greeting if configured
+            if self.conversation.greeting:
+                await self._speak(self.conversation.greeting)
+            
+        except Exception as e:
+            logger.error(f"Failed to start pipeline: {e}")
+            raise
+    
+    async def process_audio(self, pcm_bytes: bytes) -> None:
+        """
+        Process incoming audio chunk.
+        
+        This is the main entry point for audio from the user.
+        
+        Args:
+            pcm_bytes: PCM audio data (16-bit, mono, 16kHz)
+        """
+        if not self._running:
+            return
+        
+        try:
+            # 1. Process through VAD
+            vad_result = self.vad_processor.process(pcm_bytes, settings.chunk_size_ms)
+            
+            vad_status = "Silence"
+            if vad_result:
+                event_type, probability = vad_result
+                vad_status = "Speech" if event_type == "speaking" else "Silence"
+                
+                # Emit VAD event
+                await self.event_bus.publish(event_type, {
+                    "trackId": self.session_id,
+                    "probability": probability
+                })
+            else:
+                # No state change - keep previous status
+                vad_status = self._last_vad_status
+            
+            # Update state based on VAD
+            if vad_status == "Speech" and self._last_vad_status != "Speech":
+                await self._on_speech_start()
+            
+            self._last_vad_status = vad_status
+            
+            # 2. Check for barge-in (user speaking while bot speaking)
+            if self._is_bot_speaking and vad_status == "Speech":
+                await self._handle_barge_in()
+            
+            # 3. Buffer audio for ASR
+            if vad_status == "Speech" or self.conversation.state == ConversationState.LISTENING:
+                self._audio_buffer += pcm_bytes
+                await self.asr_service.send_audio(pcm_bytes)
+            
+            # 4. Check for End of Utterance
+            if self.eou_detector.process(vad_status):
+                await self._on_end_of_utterance()
+            
+        except Exception as e:
+            logger.error(f"Pipeline audio processing error: {e}", exc_info=True)
+    
+    async def process_text(self, text: str) -> None:
+        """
+        Process text input (chat command).
+        
+        Allows direct text input to bypass ASR.
+        
+        Args:
+            text: User text input
+        """
+        if not self._running:
+            return
+        
+        logger.info(f"Processing text input: {text[:50]}...")
+        
+        # Cancel any current speaking
+        await self._stop_current_speech()
+        
+        # Start new turn
+        await self.conversation.end_user_turn(text)
+        self._current_turn_task = asyncio.create_task(self._handle_turn(text))
+    
+    async def interrupt(self) -> None:
+        """Interrupt current bot speech (manual interrupt command)."""
+        await self._handle_barge_in()
+    
+    async def _on_speech_start(self) -> None:
+        """Handle user starting to speak."""
+        if self.conversation.state == ConversationState.IDLE:
+            await self.conversation.start_user_turn()
+            self._audio_buffer = b""
+            self.eou_detector.reset()
+            logger.debug("User speech started")
+    
+    async def _on_end_of_utterance(self) -> None:
+        """Handle end of user utterance."""
+        if self.conversation.state != ConversationState.LISTENING:
+            return
+        
+        # Get transcribed text (if using ASR that provides it)
+        user_text = ""
+        if hasattr(self.asr_service, 'get_and_clear_text'):
+            user_text = self.asr_service.get_and_clear_text()
+        
+        # If no ASR text, we could use the audio buffer for external ASR
+        # For now, just use placeholder if no ASR text
+        if not user_text:
+            # In a real implementation, you'd send audio_buffer to ASR here
+            # For demo purposes, use mock text
+            user_text = "[User speech detected]"
+            logger.warning("No ASR text available - using placeholder")
+        
+        logger.info(f"EOU detected - user said: {user_text[:50]}...")
+        
+        # Clear buffers
+        self._audio_buffer = b""
+        
+        # Process the turn
+        await self.conversation.end_user_turn(user_text)
+        self._current_turn_task = asyncio.create_task(self._handle_turn(user_text))
+    
+    async def _handle_turn(self, user_text: str) -> None:
+        """
+        Handle a complete conversation turn.
+        
+        Uses sentence-by-sentence streaming TTS for lower latency.
+        
+        Args:
+            user_text: User's transcribed text
+        """
+        try:
+            # Get AI response (streaming)
+            messages = self.conversation.get_messages()
+            full_response = ""
+            
+            await self.conversation.start_assistant_turn()
+            self._is_bot_speaking = True
+            self._interrupt_event.clear()
+            
+            # Sentence buffer for streaming TTS
+            sentence_buffer = ""
+            sentence_ends = {'.', '!', '?', '。', '！', '？', '；', '\n'}
+            first_audio_sent = False
+            
+            # Stream LLM response and TTS sentence by sentence
+            async for text_chunk in self.llm_service.generate_stream(messages):
+                if self._interrupt_event.is_set():
+                    break
+                
+                full_response += text_chunk
+                sentence_buffer += text_chunk
+                await self.conversation.update_assistant_text(text_chunk)
+                
+                # Check for sentence completion - synthesize immediately for low latency
+                while any(end in sentence_buffer for end in sentence_ends):
+                    # Find first sentence end
+                    min_idx = len(sentence_buffer)
+                    for end in sentence_ends:
+                        idx = sentence_buffer.find(end)
+                        if idx != -1 and idx < min_idx:
+                            min_idx = idx
+                    
+                    if min_idx < len(sentence_buffer):
+                        sentence = sentence_buffer[:min_idx + 1].strip()
+                        sentence_buffer = sentence_buffer[min_idx + 1:]
+                        
+                        if sentence and not self._interrupt_event.is_set():
+                            # Send track start on first audio
+                            if not first_audio_sent:
+                                await self.transport.send_event({
+                                    "event": "trackStart",
+                                    "trackId": self.session_id,
+                                    "timestamp": self._get_timestamp_ms()
+                                })
+                                first_audio_sent = True
+                            
+                            # Synthesize and send this sentence immediately
+                            await self._speak_sentence(sentence)
+                    else:
+                        break
+            
+            # Speak any remaining text
+            if sentence_buffer.strip() and not self._interrupt_event.is_set():
+                if not first_audio_sent:
+                    await self.transport.send_event({
+                        "event": "trackStart",
+                        "trackId": self.session_id,
+                        "timestamp": self._get_timestamp_ms()
+                    })
+                    first_audio_sent = True
+                await self._speak_sentence(sentence_buffer.strip())
+            
+            # Send track end
+            if first_audio_sent:
+                await self.transport.send_event({
+                    "event": "trackEnd",
+                    "trackId": self.session_id,
+                    "timestamp": self._get_timestamp_ms()
+                })
+            
+            # End assistant turn
+            await self.conversation.end_assistant_turn(
+                was_interrupted=self._interrupt_event.is_set()
+            )
+            
+        except asyncio.CancelledError:
+            logger.info("Turn handling cancelled")
+            await self.conversation.end_assistant_turn(was_interrupted=True)
+        except Exception as e:
+            logger.error(f"Turn handling error: {e}", exc_info=True)
+            await self.conversation.end_assistant_turn(was_interrupted=True)
+        finally:
+            self._is_bot_speaking = False
+    
+    async def _speak_sentence(self, text: str) -> None:
+        """
+        Synthesize and send a single sentence.
+        
+        Args:
+            text: Sentence to speak
+        """
+        if not text.strip() or self._interrupt_event.is_set():
+            return
+        
+        try:
+            async for chunk in self.tts_service.synthesize_stream(text):
+                if self._interrupt_event.is_set():
+                    break
+                await self.transport.send_audio(chunk.audio)
+                await asyncio.sleep(0.005)  # Small delay to prevent flooding
+        except Exception as e:
+            logger.error(f"TTS sentence error: {e}")
+    
+    async def _speak(self, text: str) -> None:
+        """
+        Synthesize and send speech.
+        
+        Args:
+            text: Text to speak
+        """
+        if not text.strip():
+            return
+        
+        try:
+            # Send track start event
+            await self.transport.send_event({
+                "event": "trackStart",
+                "trackId": self.session_id,
+                "timestamp": self._get_timestamp_ms()
+            })
+            
+            self._is_bot_speaking = True
+            
+            # Stream TTS audio
+            async for chunk in self.tts_service.synthesize_stream(text):
+                if self._interrupt_event.is_set():
+                    logger.info("TTS interrupted by barge-in")
+                    break
+                
+                # Send audio to client
+                await self.transport.send_audio(chunk.audio)
+                
+                # Small delay to prevent flooding
+                await asyncio.sleep(0.01)
+            
+            # Send track end event
+            await self.transport.send_event({
+                "event": "trackEnd",
+                "trackId": self.session_id,
+                "timestamp": self._get_timestamp_ms()
+            })
+            
+        except asyncio.CancelledError:
+            logger.info("TTS cancelled")
+            raise
+        except Exception as e:
+            logger.error(f"TTS error: {e}")
+        finally:
+            self._is_bot_speaking = False
+    
+    async def _handle_barge_in(self) -> None:
+        """Handle user barge-in (interruption)."""
+        if not self._is_bot_speaking:
+            return
+        
+        logger.info("Barge-in detected - interrupting bot speech")
+        
+        # Signal interruption
+        self._interrupt_event.set()
+        
+        # Cancel TTS
+        if self.tts_service:
+            await self.tts_service.cancel()
+        
+        # Cancel LLM
+        if self.llm_service and hasattr(self.llm_service, 'cancel'):
+            self.llm_service.cancel()
+        
+        # Interrupt conversation
+        await self.conversation.interrupt()
+        
+        # Send interrupt event to client
+        await self.transport.send_event({
+            "event": "interrupt",
+            "trackId": self.session_id,
+            "timestamp": self._get_timestamp_ms()
+        })
+        
+        # Reset for new user turn
+        self._is_bot_speaking = False
+        await self.conversation.start_user_turn()
+        self._audio_buffer = b""
+        self.eou_detector.reset()
+    
+    async def _stop_current_speech(self) -> None:
+        """Stop any current speech task."""
+        if self._current_turn_task and not self._current_turn_task.done():
+            self._interrupt_event.set()
+            self._current_turn_task.cancel()
+            try:
+                await self._current_turn_task
+            except asyncio.CancelledError:
+                pass
+        
+        self._is_bot_speaking = False
+        self._interrupt_event.clear()
+    
+    async def cleanup(self) -> None:
+        """Cleanup pipeline resources."""
+        logger.info(f"Cleaning up DuplexPipeline for session {self.session_id}")
+        
+        self._running = False
+        await self._stop_current_speech()
+        
+        # Disconnect services
+        if self.llm_service:
+            await self.llm_service.disconnect()
+        if self.tts_service:
+            await self.tts_service.disconnect()
+        if self.asr_service:
+            await self.asr_service.disconnect()
+    
+    def _get_timestamp_ms(self) -> int:
+        """Get current timestamp in milliseconds."""
+        import time
+        return int(time.time() * 1000)
+    
+    @property
+    def is_speaking(self) -> bool:
+        """Check if bot is currently speaking."""
+        return self._is_bot_speaking
+    
+    @property
+    def state(self) -> ConversationState:
+        """Get current conversation state."""
+        return self.conversation.state

core/session.py

@@ -8,6 +8,7 @@ from loguru import logger
 from core.transports import BaseTransport
 from core.pipeline import AudioPipeline
 from models.commands import parse_command, TTSCommand, ChatCommand, InterruptCommand, HangupCommand
+from app.config import settings
 
 
 class Session:
@@ -15,28 +16,44 @@ class Session:
     Manages a single call session.
 
     Handles command routing, audio processing, and session lifecycle.
+    Supports both basic audio pipeline and full duplex voice conversation.
     """
 
-    def __init__(self, session_id: str, transport: BaseTransport):
+    def __init__(self, session_id: str, transport: BaseTransport, use_duplex: bool = None):
         """
         Initialize session.
 
         Args:
             session_id: Unique session identifier
             transport: Transport instance for communication
+            use_duplex: Whether to use duplex pipeline (defaults to settings.duplex_enabled)
         """
         self.id = session_id
         self.transport = transport
+        
+        # Determine pipeline mode
+        self.use_duplex = use_duplex if use_duplex is not None else settings.duplex_enabled
+        
+        if self.use_duplex:
+            from core.duplex_pipeline import DuplexPipeline
+            self.pipeline = DuplexPipeline(
+                transport=transport,
+                session_id=session_id,
+                system_prompt=settings.duplex_system_prompt,
+                greeting=settings.duplex_greeting
+            )
+        else:
             self.pipeline = AudioPipeline(transport, session_id)
 
         # Session state
         self.created_at = None
         self.state = "created"  # created, invited, accepted, ringing, hungup
+        self._pipeline_started = False
 
         # Track IDs
         self.current_track_id: Optional[str] = str(uuid.uuid4())
 
-        logger.info(f"Session {self.id} created")
+        logger.info(f"Session {self.id} created (duplex={self.use_duplex})")
 
     async def handle_text(self, text_data: str) -> None:
         """
@@ -112,6 +129,9 @@ class Session:
             audio_bytes: PCM audio data
         """
         try:
+            if self.use_duplex:
+                await self.pipeline.process_audio(audio_bytes)
+            else:
                 await self.pipeline.process_input(audio_bytes)
         except Exception as e:
             logger.error(f"Session {self.id} handle_audio error: {e}", exc_info=True)
@@ -128,6 +148,15 @@ class Session:
             "timestamp": self._get_timestamp_ms()
         })
 
+        # Start duplex pipeline if enabled
+        if self.use_duplex and not self._pipeline_started:
+            try:
+                await self.pipeline.start()
+                self._pipeline_started = True
+                logger.info(f"Session {self.id} duplex pipeline started")
+            except Exception as e:
+                logger.error(f"Failed to start duplex pipeline: {e}")
+
         logger.info(f"Session {self.id} invited with codec: {option.get('codec', 'pcm')}")
 
     async def _handle_accept(self, data: Dict[str, Any]) -> None:
@@ -199,6 +228,9 @@ class Session:
             logger.info(f"Session {self.id} graceful interrupt")
         else:
             logger.info(f"Session {self.id} immediate interrupt")
+            if self.use_duplex:
+                await self.pipeline.interrupt()
+            else:
                 await self.pipeline.interrupt()
 
     async def _handle_pause(self) -> None:
@@ -236,6 +268,9 @@ class Session:
         """Handle chat command."""
         logger.info(f"Session {self.id} chat: {command.text[:50]}...")
         # Process text input through pipeline
+        if self.use_duplex:
+            await self.pipeline.process_text(command.text)
+        else:
             await self.pipeline.process_text_input(command.text)
 
     async def _send_error(self, sender: str, error_message: str) -> None:

examples/mic_client.py

@@ -1,137 +1,517 @@
+#!/usr/bin/env python3
 """
-Microphone WebSocket Client
+Microphone client for testing duplex voice conversation.
 
-Connects to the backend WebSocket endpoint and streams audio from the microphone.
-Used to test VAD and EOU detection.
+This client captures audio from the microphone, sends it to the server,
+and plays back the AI's voice response through the speakers.
 
-Dependencies:
-    pip install pyaudio aiohttp
+Usage:
+    python examples/mic_client.py --url ws://localhost:8000/ws
+    python examples/mic_client.py --url ws://localhost:8000/ws --chat "Hello!"
+
+Requirements:
+    pip install sounddevice soundfile websockets numpy
 """
 
+import argparse
 import asyncio
-import aiohttp
-import pyaudio
 import json
 import sys
-from datetime import datetime
+import threading
+import queue
+from pathlib import Path
 
-# Configuration
-SERVER_URL = "ws://localhost:8000/ws"
-SAMPLE_RATE = 16000
-CHANNELS = 1
-CHUNK_DURATION_MS = 20
-CHUNK_SIZE = int(SAMPLE_RATE * (CHUNK_DURATION_MS / 1000.0))  # 320 samples for 20ms
-FORMAT = pyaudio.paInt16
+try:
+    import numpy as np
+except ImportError:
+    print("Please install numpy: pip install numpy")
+    sys.exit(1)
 
-async def send_audio_loop(ws, stream):
-    """Read from microphone and send to WebSocket."""
-    print("🎙️  Microphone streaming started...")
-    try:
-        while True:
-            # Read non-blocking? PyAudio read is blocking, so run in executor or use specialized async lib.
-            # For simplicity in this script, we'll just read. It might block the event loop slightly 
-            # but for 20ms chunks it's usually acceptable for a test script.
-            # To be proper async, we should run_in_executor.
-            data = await asyncio.get_event_loop().run_in_executor(
-                None, lambda: stream.read(CHUNK_SIZE, exception_on_overflow=False)
-            )
+try:
+    import sounddevice as sd
+except ImportError:
+    print("Please install sounddevice: pip install sounddevice")
+    sys.exit(1)
 
-            await ws.send_bytes(data)
-            # No sleep needed here as microphone dictates the timing
+try:
+    import websockets
+except ImportError:
+    print("Please install websockets: pip install websockets")
+    sys.exit(1)
 
-    except Exception as e:
-        print(f"❌ Error in send loop: {e}")
 
-async def receive_loop(ws):
-    """Listen for VAD/EOU events."""
-    print("👂 Listening for server events...")
-    async for msg in ws:
-        timestamp = datetime.now().strftime("%H:%M:%S.%f")[:-3]
+class MicrophoneClient:
+    """
+    Full-duplex microphone client for voice conversation.
     
-        if msg.type == aiohttp.WSMsgType.TEXT:
-            try:
-                data = json.loads(msg.data)
-                event = data.get('event')
+    Features:
+    - Real-time microphone capture
+    - Real-time speaker playback
+    - WebSocket communication
+    - Text chat support
+    """
     
-                # Highlight VAD/EOU events
-                if event == 'speaking':
-                    print(f"[{timestamp}] 🗣️  SPEAKING STARTED")
-                elif event == 'silence':
-                    print(f"[{timestamp}] 🤫 SILENCE DETECTED")
-                elif event == 'eou':
-                    print(f"[{timestamp}] ✅ END OF UTTERANCE (EOU)")
-                elif event == 'error':
-                    print(f"[{timestamp}] ❌ ERROR: {data.get('error')}")
-                else:
-                    print(f"[{timestamp}] 📩 {event}: {str(data)[:100]}")
+    def __init__(
+        self,
+        url: str,
+        sample_rate: int = 16000,
+        chunk_duration_ms: int = 20,
+        input_device: int = None,
+        output_device: int = None
+    ):
+        """
+        Initialize microphone client.
         
-            except json.JSONDecodeError:
-                print(f"[{timestamp}] 📄 Text: {msg.data}")
+        Args:
+            url: WebSocket server URL
+            sample_rate: Audio sample rate (Hz)
+            chunk_duration_ms: Audio chunk duration (ms)
+            input_device: Input device ID (None for default)
+            output_device: Output device ID (None for default)
+        """
+        self.url = url
+        self.sample_rate = sample_rate
+        self.chunk_duration_ms = chunk_duration_ms
+        self.chunk_samples = int(sample_rate * chunk_duration_ms / 1000)
+        self.input_device = input_device
+        self.output_device = output_device
         
-        elif msg.type == aiohttp.WSMsgType.CLOSED:
-            print("❌ Connection closed")
-            break
-        elif msg.type == aiohttp.WSMsgType.ERROR:
-            print("❌ Connection error")
-            break
+        # WebSocket connection
+        self.ws = None
+        self.running = False
         
-async def main():
-    p = pyaudio.PyAudio()
+        # Audio buffers
+        self.audio_input_queue = queue.Queue()
+        self.audio_output_buffer = b""  # Continuous buffer for smooth playback
+        self.audio_output_lock = threading.Lock()
         
-    # Check for input devices
-    info = p.get_host_api_info_by_index(0)
-    numdevices = info.get('deviceCount')
-    if numdevices == 0:
-        print("❌ No audio input devices found")
-        return
+        # Statistics
+        self.bytes_sent = 0
+        self.bytes_received = 0
         
-    # Open microphone stream
-    try:
-        stream = p.open(format=FORMAT,
-                        channels=CHANNELS,
-                        rate=SAMPLE_RATE,
-                        input=True,
-                        frames_per_buffer=CHUNK_SIZE)
-    except Exception as e:
-        print(f"❌ Failed to open microphone: {e}")
-        return
+        # State
+        self.is_recording = True
+        self.is_playing = True
     
-    session = aiohttp.ClientSession()
+    async def connect(self) -> None:
+        """Connect to WebSocket server."""
+        print(f"Connecting to {self.url}...")
+        self.ws = await websockets.connect(self.url)
+        self.running = True
+        print("Connected!")
         
-    try:
-        print(f"🔌 Connecting to {SERVER_URL}...")
-        async with session.ws_connect(SERVER_URL) as ws:
-            print("✅ Connected!")
-
-            # 1. Send Invite
-            invite_msg = {
+        # Send invite command
+        await self.send_command({
             "command": "invite",
             "option": {
                 "codec": "pcm",
-                    "samplerate": SAMPLE_RATE
+                "sampleRate": self.sample_rate
             }
-            }
-            await ws.send_json(invite_msg)
-            print("📤 Sent Invite")
+        })
     
-            # 2. Run loops
-            await asyncio.gather(
-                receive_loop(ws),
-                send_audio_loop(ws, stream)
+    async def send_command(self, cmd: dict) -> None:
+        """Send JSON command to server."""
+        if self.ws:
+            await self.ws.send(json.dumps(cmd))
+            print(f"→ Command: {cmd.get('command', 'unknown')}")
+    
+    async def send_chat(self, text: str) -> None:
+        """Send chat message (text input)."""
+        await self.send_command({
+            "command": "chat",
+            "text": text
+        })
+        print(f"→ Chat: {text}")
+    
+    async def send_interrupt(self) -> None:
+        """Send interrupt command."""
+        await self.send_command({
+            "command": "interrupt"
+        })
+    
+    async def send_hangup(self, reason: str = "User quit") -> None:
+        """Send hangup command."""
+        await self.send_command({
+            "command": "hangup",
+            "reason": reason
+        })
+    
+    def _audio_input_callback(self, indata, frames, time, status):
+        """Callback for audio input (microphone)."""
+        if status:
+            print(f"Input status: {status}")
+        
+        if self.is_recording and self.running:
+            # Convert to 16-bit PCM
+            audio_data = (indata[:, 0] * 32767).astype(np.int16).tobytes()
+            self.audio_input_queue.put(audio_data)
+    
+    def _add_audio_to_buffer(self, audio_data: bytes):
+        """Add audio data to playback buffer."""
+        with self.audio_output_lock:
+            self.audio_output_buffer += audio_data
+    
+    async def _playback_task(self):
+        """Background task to play buffered audio smoothly using output stream."""
+        # Use a continuous output stream for smooth playback
+        chunk_samples = int(self.sample_rate * 0.05)  # 50ms chunks
+        chunk_bytes = chunk_samples * 2  # 16-bit = 2 bytes per sample
+        
+        def output_callback(outdata, frames, time_info, status):
+            """Audio output callback."""
+            if status:
+                print(f"Output status: {status}")
+            
+            bytes_needed = frames * 2
+            with self.audio_output_lock:
+                if len(self.audio_output_buffer) >= bytes_needed:
+                    audio_data = self.audio_output_buffer[:bytes_needed]
+                    self.audio_output_buffer = self.audio_output_buffer[bytes_needed:]
+                    samples = np.frombuffer(audio_data, dtype=np.int16).astype(np.float32) / 32767.0
+                    outdata[:, 0] = samples
+                else:
+                    outdata.fill(0)
+        
+        # Create and start output stream
+        try:
+            output_stream = sd.OutputStream(
+                samplerate=self.sample_rate,
+                channels=1,
+                dtype=np.float32,
+                blocksize=chunk_samples,
+                device=self.output_device,
+                callback=output_callback,
+                latency='low'
+            )
+            output_stream.start()
+            print(f"Audio output stream started (device: {self.output_device or 'default'})")
+            
+            # Keep stream running while client is active
+            while self.running:
+                await asyncio.sleep(0.1)
+            
+            output_stream.stop()
+            output_stream.close()
+            
+        except Exception as e:
+            print(f"Playback error: {e}")
+            import traceback
+            traceback.print_exc()
+    
+    async def audio_sender(self) -> None:
+        """Send audio from microphone to server."""
+        while self.running:
+            try:
+                # Get audio from queue with timeout
+                try:
+                    audio_data = await asyncio.get_event_loop().run_in_executor(
+                        None, lambda: self.audio_input_queue.get(timeout=0.1)
+                    )
+                except queue.Empty:
+                    continue
+                
+                # Send to server
+                if self.ws and self.is_recording:
+                    await self.ws.send(audio_data)
+                    self.bytes_sent += len(audio_data)
+                    
+            except asyncio.CancelledError:
+                break
+            except Exception as e:
+                print(f"Audio sender error: {e}")
+                break
+    
+    async def receiver(self) -> None:
+        """Receive messages from server."""
+        try:
+            while self.running:
+                try:
+                    message = await asyncio.wait_for(self.ws.recv(), timeout=0.1)
+                    
+                    if isinstance(message, bytes):
+                        # Audio data received
+                        self.bytes_received += len(message)
+                        
+                        if self.is_playing:
+                            self._add_audio_to_buffer(message)
+                        
+                        # Show progress (less verbose)
+                        with self.audio_output_lock:
+                            buffer_ms = len(self.audio_output_buffer) / (self.sample_rate * 2) * 1000
+                        duration_ms = len(message) / (self.sample_rate * 2) * 1000
+                        print(f"← Audio: {duration_ms:.0f}ms (buffer: {buffer_ms:.0f}ms)")
+                        
+                    else:
+                        # JSON event
+                        event = json.loads(message)
+                        await self._handle_event(event)
+                        
+                except asyncio.TimeoutError:
+                    continue
+                except websockets.ConnectionClosed:
+                    print("Connection closed")
+                    self.running = False
+                    break
+                    
+        except asyncio.CancelledError:
+            pass
+        except Exception as e:
+            print(f"Receiver error: {e}")
+            self.running = False
+    
+    async def _handle_event(self, event: dict) -> None:
+        """Handle incoming event."""
+        event_type = event.get("event", "unknown")
+        
+        if event_type == "answer":
+            print("← Session ready!")
+        elif event_type == "speaking":
+            print("← User speech detected")
+        elif event_type == "silence":
+            print("← User silence detected")
+        elif event_type == "trackStart":
+            print("← Bot started speaking")
+            # Clear any old audio in buffer
+            with self.audio_output_lock:
+                self.audio_output_buffer = b""
+        elif event_type == "trackEnd":
+            print("← Bot finished speaking")
+        elif event_type == "interrupt":
+            print("← Bot interrupted!")
+        elif event_type == "error":
+            print(f"← Error: {event.get('error')}")
+        elif event_type == "hangup":
+            print(f"← Hangup: {event.get('reason')}")
+            self.running = False
+        else:
+            print(f"← Event: {event_type}")
+    
+    async def interactive_mode(self) -> None:
+        """Run interactive mode for text chat."""
+        print("\n" + "=" * 50)
+        print("Voice Conversation Client")
+        print("=" * 50)
+        print("Speak into your microphone to talk to the AI.")
+        print("Or type messages to send text.")
+        print("")
+        print("Commands:")
+        print("  /quit      - End conversation")
+        print("  /mute      - Mute microphone")
+        print("  /unmute    - Unmute microphone")
+        print("  /interrupt - Interrupt AI speech")
+        print("  /stats     - Show statistics")
+        print("=" * 50 + "\n")
+        
+        while self.running:
+            try:
+                user_input = await asyncio.get_event_loop().run_in_executor(
+                    None, input, ""
                 )
                 
-    except aiohttp.ClientConnectorError:
-        print(f"❌ Failed to connect to {SERVER_URL}. Is the server running?")
-    except KeyboardInterrupt:
-        print("\n👋 Stopping...")
+                if not user_input:
+                    continue
+                
+                # Handle commands
+                if user_input.startswith("/"):
+                    cmd = user_input.lower().strip()
+                    
+                    if cmd == "/quit":
+                        await self.send_hangup("User quit")
+                        break
+                    elif cmd == "/mute":
+                        self.is_recording = False
+                        print("Microphone muted")
+                    elif cmd == "/unmute":
+                        self.is_recording = True
+                        print("Microphone unmuted")
+                    elif cmd == "/interrupt":
+                        await self.send_interrupt()
+                    elif cmd == "/stats":
+                        print(f"Sent: {self.bytes_sent / 1024:.1f} KB")
+                        print(f"Received: {self.bytes_received / 1024:.1f} KB")
+                    else:
+                        print(f"Unknown command: {cmd}")
+                else:
+                    # Send as chat message
+                    await self.send_chat(user_input)
+                    
+            except EOFError:
+                break
+            except Exception as e:
+                print(f"Input error: {e}")
+    
+    async def run(self, chat_message: str = None, interactive: bool = True) -> None:
+        """
+        Run the client.
+        
+        Args:
+            chat_message: Optional single chat message to send
+            interactive: Whether to run in interactive mode
+        """
+        try:
+            await self.connect()
+            
+            # Wait for answer
+            await asyncio.sleep(0.5)
+            
+            # Start audio input stream
+            print("Starting audio streams...")
+            
+            input_stream = sd.InputStream(
+                samplerate=self.sample_rate,
+                channels=1,
+                dtype=np.float32,
+                blocksize=self.chunk_samples,
+                device=self.input_device,
+                callback=self._audio_input_callback
+            )
+            
+            input_stream.start()
+            print("Audio streams started")
+            
+            # Start background tasks
+            sender_task = asyncio.create_task(self.audio_sender())
+            receiver_task = asyncio.create_task(self.receiver())
+            playback_task = asyncio.create_task(self._playback_task())
+            
+            if chat_message:
+                # Send single message and wait
+                await self.send_chat(chat_message)
+                await asyncio.sleep(15)
+            elif interactive:
+                # Run interactive mode
+                await self.interactive_mode()
+            else:
+                # Just wait
+                while self.running:
+                    await asyncio.sleep(0.1)
+            
+            # Cleanup
+            self.running = False
+            sender_task.cancel()
+            receiver_task.cancel()
+            playback_task.cancel()
+            
+            try:
+                await sender_task
+            except asyncio.CancelledError:
+                pass
+            
+            try:
+                await receiver_task
+            except asyncio.CancelledError:
+                pass
+            
+            try:
+                await playback_task
+            except asyncio.CancelledError:
+                pass
+            
+            input_stream.stop()
+            
+        except ConnectionRefusedError:
+            print(f"Error: Could not connect to {self.url}")
+            print("Make sure the server is running.")
+        except Exception as e:
+            print(f"Error: {e}")
         finally:
-        stream.stop_stream()
-        stream.close()
-        p.terminate()
-        await session.close()
+            await self.close()
+    
+    async def close(self) -> None:
+        """Close the connection."""
+        self.running = False
+        if self.ws:
+            await self.ws.close()
+        
+        print(f"\nSession ended")
+        print(f"  Total sent: {self.bytes_sent / 1024:.1f} KB")
+        print(f"  Total received: {self.bytes_received / 1024:.1f} KB")
+
+
+def list_devices():
+    """List available audio devices."""
+    print("\nAvailable audio devices:")
+    print("-" * 60)
+    devices = sd.query_devices()
+    for i, device in enumerate(devices):
+        direction = []
+        if device['max_input_channels'] > 0:
+            direction.append("IN")
+        if device['max_output_channels'] > 0:
+            direction.append("OUT")
+        direction_str = "/".join(direction) if direction else "N/A"
+        
+        default = ""
+        if i == sd.default.device[0]:
+            default += " [DEFAULT INPUT]"
+        if i == sd.default.device[1]:
+            default += " [DEFAULT OUTPUT]"
+        
+        print(f"  {i:2d}: {device['name'][:40]:40s} ({direction_str}){default}")
+    print("-" * 60)
+
+
+async def main():
+    parser = argparse.ArgumentParser(
+        description="Microphone client for duplex voice conversation"
+    )
+    parser.add_argument(
+        "--url",
+        default="ws://localhost:8000/ws",
+        help="WebSocket server URL"
+    )
+    parser.add_argument(
+        "--chat",
+        help="Send a single chat message instead of using microphone"
+    )
+    parser.add_argument(
+        "--sample-rate",
+        type=int,
+        default=16000,
+        help="Audio sample rate (default: 16000)"
+    )
+    parser.add_argument(
+        "--input-device",
+        type=int,
+        help="Input device ID"
+    )
+    parser.add_argument(
+        "--output-device",
+        type=int,
+        help="Output device ID"
+    )
+    parser.add_argument(
+        "--list-devices",
+        action="store_true",
+        help="List available audio devices and exit"
+    )
+    parser.add_argument(
+        "--no-interactive",
+        action="store_true",
+        help="Disable interactive mode"
+    )
+    
+    args = parser.parse_args()
+    
+    if args.list_devices:
+        list_devices()
+        return
+    
+    client = MicrophoneClient(
+        url=args.url,
+        sample_rate=args.sample_rate,
+        input_device=args.input_device,
+        output_device=args.output_device
+    )
+    
+    await client.run(
+        chat_message=args.chat,
+        interactive=not args.no_interactive
+    )
+
 
 if __name__ == "__main__":
     try:
         asyncio.run(main())
     except KeyboardInterrupt:
-        pass
+        print("\nInterrupted by user")

examples/simple_client.py

@@ -0,0 +1,239 @@
+#!/usr/bin/env python3
+"""
+Simple WebSocket client for testing voice conversation.
+Uses PyAudio for more reliable audio playback on Windows.
+
+Usage:
+    python examples/simple_client.py
+    python examples/simple_client.py --text "Hello"
+"""
+
+import argparse
+import asyncio
+import json
+import sys
+import wave
+import io
+
+try:
+    import numpy as np
+except ImportError:
+    print("pip install numpy")
+    sys.exit(1)
+
+try:
+    import websockets
+except ImportError:
+    print("pip install websockets")
+    sys.exit(1)
+
+# Try PyAudio first (more reliable on Windows)
+try:
+    import pyaudio
+    PYAUDIO_AVAILABLE = True
+except ImportError:
+    PYAUDIO_AVAILABLE = False
+    print("PyAudio not available, trying sounddevice...")
+
+try:
+    import sounddevice as sd
+    SD_AVAILABLE = True
+except ImportError:
+    SD_AVAILABLE = False
+
+if not PYAUDIO_AVAILABLE and not SD_AVAILABLE:
+    print("Please install pyaudio or sounddevice:")
+    print("  pip install pyaudio")
+    print("  or: pip install sounddevice")
+    sys.exit(1)
+
+
+class SimpleVoiceClient:
+    """Simple voice client with reliable audio playback."""
+    
+    def __init__(self, url: str, sample_rate: int = 16000):
+        self.url = url
+        self.sample_rate = sample_rate
+        self.ws = None
+        self.running = False
+        
+        # Audio buffer
+        self.audio_buffer = b""
+        
+        # PyAudio setup
+        if PYAUDIO_AVAILABLE:
+            self.pa = pyaudio.PyAudio()
+            self.stream = None
+        
+        # Stats
+        self.bytes_received = 0
+    
+    async def connect(self):
+        """Connect to server."""
+        print(f"Connecting to {self.url}...")
+        self.ws = await websockets.connect(self.url)
+        self.running = True
+        print("Connected!")
+        
+        # Send invite
+        await self.ws.send(json.dumps({
+            "command": "invite",
+            "option": {"codec": "pcm", "sampleRate": self.sample_rate}
+        }))
+        print("-> invite")
+    
+    async def send_chat(self, text: str):
+        """Send chat message."""
+        await self.ws.send(json.dumps({"command": "chat", "text": text}))
+        print(f"-> chat: {text}")
+    
+    def play_audio(self, audio_data: bytes):
+        """Play audio data immediately."""
+        if len(audio_data) == 0:
+            return
+            
+        if PYAUDIO_AVAILABLE:
+            # Use PyAudio - more reliable on Windows
+            if self.stream is None:
+                self.stream = self.pa.open(
+                    format=pyaudio.paInt16,
+                    channels=1,
+                    rate=self.sample_rate,
+                    output=True,
+                    frames_per_buffer=1024
+                )
+            self.stream.write(audio_data)
+        elif SD_AVAILABLE:
+            # Use sounddevice
+            samples = np.frombuffer(audio_data, dtype=np.int16).astype(np.float32) / 32767.0
+            sd.play(samples, self.sample_rate, blocking=True)
+    
+    async def receive_loop(self):
+        """Receive and play audio."""
+        print("\nWaiting for response...")
+        
+        while self.running:
+            try:
+                msg = await asyncio.wait_for(self.ws.recv(), timeout=0.1)
+                
+                if isinstance(msg, bytes):
+                    # Audio data
+                    self.bytes_received += len(msg)
+                    duration_ms = len(msg) / (self.sample_rate * 2) * 1000
+                    print(f"<- audio: {len(msg)} bytes ({duration_ms:.0f}ms)")
+                    
+                    # Play immediately in executor to not block
+                    loop = asyncio.get_event_loop()
+                    await loop.run_in_executor(None, self.play_audio, msg)
+                else:
+                    # JSON event
+                    event = json.loads(msg)
+                    etype = event.get("event", "?")
+                    print(f"<- {etype}")
+                    
+                    if etype == "hangup":
+                        self.running = False
+                        break
+                        
+            except asyncio.TimeoutError:
+                continue
+            except websockets.ConnectionClosed:
+                print("Connection closed")
+                self.running = False
+                break
+    
+    async def run(self, text: str = None):
+        """Run the client."""
+        try:
+            await self.connect()
+            await asyncio.sleep(0.5)
+            
+            # Start receiver
+            recv_task = asyncio.create_task(self.receive_loop())
+            
+            if text:
+                await self.send_chat(text)
+                # Wait for response
+                await asyncio.sleep(30)
+            else:
+                # Interactive mode
+                print("\nType a message and press Enter (or 'quit' to exit):")
+                while self.running:
+                    try:
+                        user_input = await asyncio.get_event_loop().run_in_executor(
+                            None, input, "> "
+                        )
+                        if user_input.lower() == 'quit':
+                            break
+                        if user_input.strip():
+                            await self.send_chat(user_input)
+                    except EOFError:
+                        break
+            
+            self.running = False
+            recv_task.cancel()
+            try:
+                await recv_task
+            except asyncio.CancelledError:
+                pass
+                
+        finally:
+            await self.close()
+    
+    async def close(self):
+        """Close connections."""
+        self.running = False
+        
+        if PYAUDIO_AVAILABLE:
+            if self.stream:
+                self.stream.stop_stream()
+                self.stream.close()
+            self.pa.terminate()
+        
+        if self.ws:
+            await self.ws.close()
+        
+        print(f"\nTotal audio received: {self.bytes_received / 1024:.1f} KB")
+
+
+def list_audio_devices():
+    """List available audio devices."""
+    print("\n=== Audio Devices ===")
+    
+    if PYAUDIO_AVAILABLE:
+        pa = pyaudio.PyAudio()
+        print("\nPyAudio devices:")
+        for i in range(pa.get_device_count()):
+            info = pa.get_device_info_by_index(i)
+            if info['maxOutputChannels'] > 0:
+                default = " [DEFAULT]" if i == pa.get_default_output_device_info()['index'] else ""
+                print(f"  {i}: {info['name']}{default}")
+        pa.terminate()
+    
+    if SD_AVAILABLE:
+        print("\nSounddevice devices:")
+        for i, d in enumerate(sd.query_devices()):
+            if d['max_output_channels'] > 0:
+                default = " [DEFAULT]" if i == sd.default.device[1] else ""
+                print(f"  {i}: {d['name']}{default}")
+
+
+async def main():
+    parser = argparse.ArgumentParser(description="Simple voice client")
+    parser.add_argument("--url", default="ws://localhost:8000/ws")
+    parser.add_argument("--text", help="Send text and play response")
+    parser.add_argument("--list-devices", action="store_true")
+    parser.add_argument("--sample-rate", type=int, default=16000)
+    
+    args = parser.parse_args()
+    
+    if args.list_devices:
+        list_audio_devices()
+        return
+    
+    client = SimpleVoiceClient(args.url, args.sample_rate)
+    await client.run(args.text)
+
+
+if __name__ == "__main__":
+    asyncio.run(main())

requirements.txt

@@ -0,0 +1,37 @@
+# Web Framework
+fastapi>=0.109.0
+uvicorn[standard]>=0.27.0
+websockets>=12.0
+python-multipart>=0.0.6
+
+# WebRTC (optional - for WebRTC transport)
+aiortc>=1.6.0
+
+# Audio Processing
+av>=12.1.0
+numpy>=1.26.3
+onnxruntime>=1.16.3
+
+# Configuration
+pydantic>=2.5.3
+pydantic-settings>=2.1.0
+python-dotenv>=1.0.0
+toml>=0.10.2
+
+# Logging
+loguru>=0.7.2
+
+# HTTP Client
+aiohttp>=3.9.1
+
+# AI Services - LLM
+openai>=1.0.0
+
+# AI Services - TTS
+edge-tts>=6.1.0
+pydub>=0.25.0  # For audio format conversion
+
+# Microphone client dependencies
+sounddevice>=0.4.6
+soundfile>=0.12.1
+pyaudio>=0.2.13  # More reliable audio on Windows

services/__init__.py

@@ -0,0 +1,42 @@
+"""AI Services package.
+
+Provides ASR, LLM, TTS, and Realtime API services for voice conversation.
+"""
+
+from services.base import (
+    ServiceState,
+    ASRResult,
+    LLMMessage,
+    TTSChunk,
+    BaseASRService,
+    BaseLLMService,
+    BaseTTSService,
+)
+from services.llm import OpenAILLMService, MockLLMService
+from services.tts import EdgeTTSService, MockTTSService
+from services.asr import BufferedASRService, MockASRService
+from services.realtime import RealtimeService, RealtimeConfig, RealtimePipeline
+
+__all__ = [
+    # Base classes
+    "ServiceState",
+    "ASRResult",
+    "LLMMessage",
+    "TTSChunk",
+    "BaseASRService",
+    "BaseLLMService",
+    "BaseTTSService",
+    # LLM
+    "OpenAILLMService",
+    "MockLLMService",
+    # TTS
+    "EdgeTTSService",
+    "MockTTSService",
+    # ASR
+    "BufferedASRService",
+    "MockASRService",
+    # Realtime
+    "RealtimeService",
+    "RealtimeConfig",
+    "RealtimePipeline",
+]

services/asr.py

@@ -0,0 +1,147 @@
+"""ASR (Automatic Speech Recognition) Service implementations.
+
+Provides speech-to-text capabilities with streaming support.
+"""
+
+import os
+import asyncio
+import json
+from typing import AsyncIterator, Optional
+from loguru import logger
+
+from services.base import BaseASRService, ASRResult, ServiceState
+
+# Try to import websockets for streaming ASR
+try:
+    import websockets
+    WEBSOCKETS_AVAILABLE = True
+except ImportError:
+    WEBSOCKETS_AVAILABLE = False
+
+
+class BufferedASRService(BaseASRService):
+    """
+    Buffered ASR service that accumulates audio and provides
+    a simple text accumulator for use with EOU detection.
+    
+    This is a lightweight implementation that works with the
+    existing VAD + EOU pattern without requiring external ASR.
+    """
+    
+    def __init__(
+        self,
+        sample_rate: int = 16000,
+        language: str = "en"
+    ):
+        super().__init__(sample_rate=sample_rate, language=language)
+        
+        self._audio_buffer: bytes = b""
+        self._current_text: str = ""
+        self._transcript_queue: asyncio.Queue[ASRResult] = asyncio.Queue()
+    
+    async def connect(self) -> None:
+        """No connection needed for buffered ASR."""
+        self.state = ServiceState.CONNECTED
+        logger.info("Buffered ASR service connected")
+    
+    async def disconnect(self) -> None:
+        """Clear buffers on disconnect."""
+        self._audio_buffer = b""
+        self._current_text = ""
+        self.state = ServiceState.DISCONNECTED
+        logger.info("Buffered ASR service disconnected")
+    
+    async def send_audio(self, audio: bytes) -> None:
+        """Buffer audio for later processing."""
+        self._audio_buffer += audio
+    
+    async def receive_transcripts(self) -> AsyncIterator[ASRResult]:
+        """Yield transcription results."""
+        while True:
+            try:
+                result = await asyncio.wait_for(
+                    self._transcript_queue.get(),
+                    timeout=0.1
+                )
+                yield result
+            except asyncio.TimeoutError:
+                continue
+            except asyncio.CancelledError:
+                break
+    
+    def set_text(self, text: str) -> None:
+        """
+        Set the current transcript text directly.
+        
+        This allows external integration (e.g., Whisper, other ASR)
+        to provide transcripts.
+        """
+        self._current_text = text
+        result = ASRResult(text=text, is_final=False)
+        asyncio.create_task(self._transcript_queue.put(result))
+    
+    def get_and_clear_text(self) -> str:
+        """Get accumulated text and clear buffer."""
+        text = self._current_text
+        self._current_text = ""
+        self._audio_buffer = b""
+        return text
+    
+    def get_audio_buffer(self) -> bytes:
+        """Get accumulated audio buffer."""
+        return self._audio_buffer
+    
+    def clear_audio_buffer(self) -> None:
+        """Clear audio buffer."""
+        self._audio_buffer = b""
+
+
+class MockASRService(BaseASRService):
+    """
+    Mock ASR service for testing without actual recognition.
+    """
+    
+    def __init__(self, sample_rate: int = 16000, language: str = "en"):
+        super().__init__(sample_rate=sample_rate, language=language)
+        self._transcript_queue: asyncio.Queue[ASRResult] = asyncio.Queue()
+        self._mock_texts = [
+            "Hello, how are you?",
+            "That's interesting.",
+            "Tell me more about that.",
+            "I understand.",
+        ]
+        self._text_index = 0
+    
+    async def connect(self) -> None:
+        self.state = ServiceState.CONNECTED
+        logger.info("Mock ASR service connected")
+    
+    async def disconnect(self) -> None:
+        self.state = ServiceState.DISCONNECTED
+        logger.info("Mock ASR service disconnected")
+    
+    async def send_audio(self, audio: bytes) -> None:
+        """Mock audio processing - generates fake transcripts periodically."""
+        pass
+    
+    def trigger_transcript(self) -> None:
+        """Manually trigger a transcript (for testing)."""
+        text = self._mock_texts[self._text_index % len(self._mock_texts)]
+        self._text_index += 1
+        
+        result = ASRResult(text=text, is_final=True, confidence=0.95)
+        asyncio.create_task(self._transcript_queue.put(result))
+    
+    async def receive_transcripts(self) -> AsyncIterator[ASRResult]:
+        """Yield transcription results."""
+        while True:
+            try:
+                result = await asyncio.wait_for(
+                    self._transcript_queue.get(),
+                    timeout=0.1
+                )
+                yield result
+            except asyncio.TimeoutError:
+                continue
+            except asyncio.CancelledError:
+                break

services/base.py

@@ -0,0 +1,244 @@
+"""Base classes for AI services.
+
+Defines abstract interfaces for ASR, LLM, and TTS services,
+inspired by pipecat's service architecture and active-call's
+StreamEngine pattern.
+"""
+
+from abc import ABC, abstractmethod
+from dataclasses import dataclass, field
+from typing import AsyncIterator, Optional, List, Dict, Any
+from enum import Enum
+
+
+class ServiceState(Enum):
+    """Service connection state."""
+    DISCONNECTED = "disconnected"
+    CONNECTING = "connecting"
+    CONNECTED = "connected"
+    ERROR = "error"
+
+
+@dataclass
+class ASRResult:
+    """ASR transcription result."""
+    text: str
+    is_final: bool = False
+    confidence: float = 1.0
+    language: Optional[str] = None
+    start_time: Optional[float] = None
+    end_time: Optional[float] = None
+    
+    def __str__(self) -> str:
+        status = "FINAL" if self.is_final else "PARTIAL"
+        return f"[{status}] {self.text}"
+
+
+@dataclass
+class LLMMessage:
+    """LLM conversation message."""
+    role: str  # "system", "user", "assistant", "function"
+    content: str
+    name: Optional[str] = None  # For function calls
+    function_call: Optional[Dict[str, Any]] = None
+    
+    def to_dict(self) -> Dict[str, Any]:
+        """Convert to API-compatible dict."""
+        d = {"role": self.role, "content": self.content}
+        if self.name:
+            d["name"] = self.name
+        if self.function_call:
+            d["function_call"] = self.function_call
+        return d
+
+
+@dataclass
+class TTSChunk:
+    """TTS audio chunk."""
+    audio: bytes  # PCM audio data
+    sample_rate: int = 16000
+    channels: int = 1
+    bits_per_sample: int = 16
+    is_final: bool = False
+    text_offset: Optional[int] = None  # Character offset in original text
+
+
+class BaseASRService(ABC):
+    """
+    Abstract base class for ASR (Speech-to-Text) services.
+    
+    Supports both streaming and non-streaming transcription.
+    """
+    
+    def __init__(self, sample_rate: int = 16000, language: str = "en"):
+        self.sample_rate = sample_rate
+        self.language = language
+        self.state = ServiceState.DISCONNECTED
+    
+    @abstractmethod
+    async def connect(self) -> None:
+        """Establish connection to ASR service."""
+        pass
+    
+    @abstractmethod
+    async def disconnect(self) -> None:
+        """Close connection to ASR service."""
+        pass
+    
+    @abstractmethod
+    async def send_audio(self, audio: bytes) -> None:
+        """
+        Send audio chunk for transcription.
+        
+        Args:
+            audio: PCM audio data (16-bit, mono)
+        """
+        pass
+    
+    @abstractmethod
+    async def receive_transcripts(self) -> AsyncIterator[ASRResult]:
+        """
+        Receive transcription results.
+        
+        Yields:
+            ASRResult objects as they become available
+        """
+        pass
+    
+    async def transcribe(self, audio: bytes) -> ASRResult:
+        """
+        Transcribe a complete audio buffer (non-streaming).
+        
+        Args:
+            audio: Complete PCM audio data
+            
+        Returns:
+            Final ASRResult
+        """
+        # Default implementation using streaming
+        await self.send_audio(audio)
+        async for result in self.receive_transcripts():
+            if result.is_final:
+                return result
+        return ASRResult(text="", is_final=True)
+
+
+class BaseLLMService(ABC):
+    """
+    Abstract base class for LLM (Language Model) services.
+    
+    Supports streaming responses for real-time conversation.
+    """
+    
+    def __init__(self, model: str = "gpt-4"):
+        self.model = model
+        self.state = ServiceState.DISCONNECTED
+    
+    @abstractmethod
+    async def connect(self) -> None:
+        """Initialize LLM service connection."""
+        pass
+    
+    @abstractmethod
+    async def disconnect(self) -> None:
+        """Close LLM service connection."""
+        pass
+    
+    @abstractmethod
+    async def generate(
+        self,
+        messages: List[LLMMessage],
+        temperature: float = 0.7,
+        max_tokens: Optional[int] = None
+    ) -> str:
+        """
+        Generate a complete response.
+        
+        Args:
+            messages: Conversation history
+            temperature: Sampling temperature
+            max_tokens: Maximum tokens to generate
+            
+        Returns:
+            Complete response text
+        """
+        pass
+    
+    @abstractmethod
+    async def generate_stream(
+        self,
+        messages: List[LLMMessage],
+        temperature: float = 0.7,
+        max_tokens: Optional[int] = None
+    ) -> AsyncIterator[str]:
+        """
+        Generate response in streaming mode.
+        
+        Args:
+            messages: Conversation history
+            temperature: Sampling temperature
+            max_tokens: Maximum tokens to generate
+            
+        Yields:
+            Text chunks as they are generated
+        """
+        pass
+
+
+class BaseTTSService(ABC):
+    """
+    Abstract base class for TTS (Text-to-Speech) services.
+    
+    Supports streaming audio synthesis for low-latency playback.
+    """
+    
+    def __init__(
+        self,
+        voice: str = "default",
+        sample_rate: int = 16000,
+        speed: float = 1.0
+    ):
+        self.voice = voice
+        self.sample_rate = sample_rate
+        self.speed = speed
+        self.state = ServiceState.DISCONNECTED
+    
+    @abstractmethod
+    async def connect(self) -> None:
+        """Initialize TTS service connection."""
+        pass
+    
+    @abstractmethod
+    async def disconnect(self) -> None:
+        """Close TTS service connection."""
+        pass
+    
+    @abstractmethod
+    async def synthesize(self, text: str) -> bytes:
+        """
+        Synthesize complete audio for text (non-streaming).
+        
+        Args:
+            text: Text to synthesize
+            
+        Returns:
+            Complete PCM audio data
+        """
+        pass
+    
+    @abstractmethod
+    async def synthesize_stream(self, text: str) -> AsyncIterator[TTSChunk]:
+        """
+        Synthesize audio in streaming mode.
+        
+        Args:
+            text: Text to synthesize
+            
+        Yields:
+            TTSChunk objects as audio is generated
+        """
+        pass
+    
+    async def cancel(self) -> None:
+        """Cancel ongoing synthesis (for barge-in support)."""
+        pass

services/llm.py

@@ -0,0 +1,239 @@
+"""LLM (Large Language Model) Service implementations.
+
+Provides OpenAI-compatible LLM integration with streaming support
+for real-time voice conversation.
+"""
+
+import os
+import asyncio
+from typing import AsyncIterator, Optional, List, Dict, Any
+from loguru import logger
+
+from services.base import BaseLLMService, LLMMessage, ServiceState
+
+# Try to import openai
+try:
+    from openai import AsyncOpenAI
+    OPENAI_AVAILABLE = True
+except ImportError:
+    OPENAI_AVAILABLE = False
+    logger.warning("openai package not available - LLM service will be disabled")
+
+
+class OpenAILLMService(BaseLLMService):
+    """
+    OpenAI-compatible LLM service.
+    
+    Supports streaming responses for low-latency voice conversation.
+    Works with OpenAI API, Azure OpenAI, and compatible APIs.
+    """
+    
+    def __init__(
+        self,
+        model: str = "gpt-4o-mini",
+        api_key: Optional[str] = None,
+        base_url: Optional[str] = None,
+        system_prompt: Optional[str] = None
+    ):
+        """
+        Initialize OpenAI LLM service.
+        
+        Args:
+            model: Model name (e.g., "gpt-4o-mini", "gpt-4o")
+            api_key: OpenAI API key (defaults to OPENAI_API_KEY env var)
+            base_url: Custom API base URL (for Azure or compatible APIs)
+            system_prompt: Default system prompt for conversations
+        """
+        super().__init__(model=model)
+        
+        self.api_key = api_key or os.getenv("OPENAI_API_KEY")
+        self.base_url = base_url or os.getenv("OPENAI_API_URL")
+        self.system_prompt = system_prompt or (
+            "You are a helpful, friendly voice assistant. "
+            "Keep your responses concise and conversational. "
+            "Respond naturally as if having a phone conversation."
+        )
+        
+        self.client: Optional[AsyncOpenAI] = None
+        self._cancel_event = asyncio.Event()
+    
+    async def connect(self) -> None:
+        """Initialize OpenAI client."""
+        if not OPENAI_AVAILABLE:
+            raise RuntimeError("openai package not installed")
+        
+        if not self.api_key:
+            raise ValueError("OpenAI API key not provided")
+        
+        self.client = AsyncOpenAI(
+            api_key=self.api_key,
+            base_url=self.base_url
+        )
+        self.state = ServiceState.CONNECTED
+        logger.info(f"OpenAI LLM service connected: model={self.model}")
+    
+    async def disconnect(self) -> None:
+        """Close OpenAI client."""
+        if self.client:
+            await self.client.close()
+            self.client = None
+        self.state = ServiceState.DISCONNECTED
+        logger.info("OpenAI LLM service disconnected")
+    
+    def _prepare_messages(self, messages: List[LLMMessage]) -> List[Dict[str, Any]]:
+        """Prepare messages list with system prompt."""
+        result = []
+        
+        # Add system prompt if not already present
+        has_system = any(m.role == "system" for m in messages)
+        if not has_system and self.system_prompt:
+            result.append({"role": "system", "content": self.system_prompt})
+        
+        # Add all messages
+        for msg in messages:
+            result.append(msg.to_dict())
+        
+        return result
+    
+    async def generate(
+        self,
+        messages: List[LLMMessage],
+        temperature: float = 0.7,
+        max_tokens: Optional[int] = None
+    ) -> str:
+        """
+        Generate a complete response.
+        
+        Args:
+            messages: Conversation history
+            temperature: Sampling temperature
+            max_tokens: Maximum tokens to generate
+            
+        Returns:
+            Complete response text
+        """
+        if not self.client:
+            raise RuntimeError("LLM service not connected")
+        
+        prepared = self._prepare_messages(messages)
+        
+        try:
+            response = await self.client.chat.completions.create(
+                model=self.model,
+                messages=prepared,
+                temperature=temperature,
+                max_tokens=max_tokens
+            )
+            
+            content = response.choices[0].message.content or ""
+            logger.debug(f"LLM response: {content[:100]}...")
+            return content
+            
+        except Exception as e:
+            logger.error(f"LLM generation error: {e}")
+            raise
+    
+    async def generate_stream(
+        self,
+        messages: List[LLMMessage],
+        temperature: float = 0.7,
+        max_tokens: Optional[int] = None
+    ) -> AsyncIterator[str]:
+        """
+        Generate response in streaming mode.
+        
+        Args:
+            messages: Conversation history
+            temperature: Sampling temperature
+            max_tokens: Maximum tokens to generate
+            
+        Yields:
+            Text chunks as they are generated
+        """
+        if not self.client:
+            raise RuntimeError("LLM service not connected")
+        
+        prepared = self._prepare_messages(messages)
+        self._cancel_event.clear()
+        
+        try:
+            stream = await self.client.chat.completions.create(
+                model=self.model,
+                messages=prepared,
+                temperature=temperature,
+                max_tokens=max_tokens,
+                stream=True
+            )
+            
+            async for chunk in stream:
+                # Check for cancellation
+                if self._cancel_event.is_set():
+                    logger.info("LLM stream cancelled")
+                    break
+                
+                if chunk.choices and chunk.choices[0].delta.content:
+                    content = chunk.choices[0].delta.content
+                    yield content
+            
+        except asyncio.CancelledError:
+            logger.info("LLM stream cancelled via asyncio")
+            raise
+        except Exception as e:
+            logger.error(f"LLM streaming error: {e}")
+            raise
+    
+    def cancel(self) -> None:
+        """Cancel ongoing generation."""
+        self._cancel_event.set()
+
+
+class MockLLMService(BaseLLMService):
+    """
+    Mock LLM service for testing without API calls.
+    """
+    
+    def __init__(self, response_delay: float = 0.5):
+        super().__init__(model="mock")
+        self.response_delay = response_delay
+        self.responses = [
+            "Hello! How can I help you today?",
+            "That's an interesting question. Let me think about it.",
+            "I understand. Is there anything else you'd like to know?",
+            "Great! I'm here if you need anything else.",
+        ]
+        self._response_index = 0
+    
+    async def connect(self) -> None:
+        self.state = ServiceState.CONNECTED
+        logger.info("Mock LLM service connected")
+    
+    async def disconnect(self) -> None:
+        self.state = ServiceState.DISCONNECTED
+        logger.info("Mock LLM service disconnected")
+    
+    async def generate(
+        self,
+        messages: List[LLMMessage],
+        temperature: float = 0.7,
+        max_tokens: Optional[int] = None
+    ) -> str:
+        await asyncio.sleep(self.response_delay)
+        response = self.responses[self._response_index % len(self.responses)]
+        self._response_index += 1
+        return response
+    
+    async def generate_stream(
+        self,
+        messages: List[LLMMessage],
+        temperature: float = 0.7,
+        max_tokens: Optional[int] = None
+    ) -> AsyncIterator[str]:
+        response = await self.generate(messages, temperature, max_tokens)
+        
+        # Stream word by word
+        words = response.split()
+        for i, word in enumerate(words):
+            if i > 0:
+                yield " "
+            yield word
+            await asyncio.sleep(0.05)  # Simulate streaming delay

services/realtime.py

@@ -0,0 +1,548 @@
+"""OpenAI Realtime API Service.
+
+Provides true duplex voice conversation using OpenAI's Realtime API,
+similar to active-call's RealtimeProcessor. This bypasses the need for
+separate ASR/LLM/TTS services by handling everything server-side.
+
+The Realtime API provides:
+- Server-side VAD with turn detection
+- Streaming speech-to-text
+- Streaming LLM responses
+- Streaming text-to-speech
+- Function calling support
+- Barge-in/interruption handling
+"""
+
+import os
+import asyncio
+import json
+import base64
+from typing import Optional, Dict, Any, Callable, Awaitable, List
+from dataclasses import dataclass, field
+from enum import Enum
+from loguru import logger
+
+try:
+    import websockets
+    WEBSOCKETS_AVAILABLE = True
+except ImportError:
+    WEBSOCKETS_AVAILABLE = False
+    logger.warning("websockets not available - Realtime API will be disabled")
+
+
+class RealtimeState(Enum):
+    """Realtime API connection state."""
+    DISCONNECTED = "disconnected"
+    CONNECTING = "connecting"
+    CONNECTED = "connected"
+    ERROR = "error"
+
+
+@dataclass
+class RealtimeConfig:
+    """Configuration for OpenAI Realtime API."""
+    
+    # API Configuration
+    api_key: Optional[str] = None
+    model: str = "gpt-4o-realtime-preview"
+    endpoint: Optional[str] = None  # For Azure or custom endpoints
+    
+    # Voice Configuration
+    voice: str = "alloy"  # alloy, echo, shimmer, etc.
+    instructions: str = (
+        "You are a helpful, friendly voice assistant. "
+        "Keep your responses concise and conversational."
+    )
+    
+    # Turn Detection (Server-side VAD)
+    turn_detection: Optional[Dict[str, Any]] = field(default_factory=lambda: {
+        "type": "server_vad",
+        "threshold": 0.5,
+        "prefix_padding_ms": 300,
+        "silence_duration_ms": 500
+    })
+    
+    # Audio Configuration
+    input_audio_format: str = "pcm16"
+    output_audio_format: str = "pcm16"
+    
+    # Tools/Functions
+    tools: List[Dict[str, Any]] = field(default_factory=list)
+
+
+class RealtimeService:
+    """
+    OpenAI Realtime API service for true duplex voice conversation.
+    
+    This service handles the entire voice conversation pipeline:
+    1. Audio input → Server-side VAD → Speech-to-text
+    2. Text → LLM processing → Response generation
+    3. Response → Text-to-speech → Audio output
+    
+    Events emitted:
+    - on_audio: Audio output from the assistant
+    - on_transcript: Text transcript (user or assistant)
+    - on_speech_started: User started speaking
+    - on_speech_stopped: User stopped speaking
+    - on_response_started: Assistant started responding
+    - on_response_done: Assistant finished responding
+    - on_function_call: Function call requested
+    - on_error: Error occurred
+    """
+    
+    def __init__(self, config: Optional[RealtimeConfig] = None):
+        """
+        Initialize Realtime API service.
+        
+        Args:
+            config: Realtime configuration (uses defaults if not provided)
+        """
+        self.config = config or RealtimeConfig()
+        self.config.api_key = self.config.api_key or os.getenv("OPENAI_API_KEY")
+        
+        self.state = RealtimeState.DISCONNECTED
+        self._ws = None
+        self._receive_task: Optional[asyncio.Task] = None
+        self._cancel_event = asyncio.Event()
+        
+        # Event callbacks
+        self._callbacks: Dict[str, List[Callable]] = {
+            "on_audio": [],
+            "on_transcript": [],
+            "on_speech_started": [],
+            "on_speech_stopped": [],
+            "on_response_started": [],
+            "on_response_done": [],
+            "on_function_call": [],
+            "on_error": [],
+            "on_interrupted": [],
+        }
+        
+        logger.debug(f"RealtimeService initialized with model={self.config.model}")
+    
+    def on(self, event: str, callback: Callable[..., Awaitable[None]]) -> None:
+        """
+        Register event callback.
+        
+        Args:
+            event: Event name
+            callback: Async callback function
+        """
+        if event in self._callbacks:
+            self._callbacks[event].append(callback)
+    
+    async def _emit(self, event: str, *args, **kwargs) -> None:
+        """Emit event to all registered callbacks."""
+        for callback in self._callbacks.get(event, []):
+            try:
+                await callback(*args, **kwargs)
+            except Exception as e:
+                logger.error(f"Event callback error ({event}): {e}")
+    
+    async def connect(self) -> None:
+        """Connect to OpenAI Realtime API."""
+        if not WEBSOCKETS_AVAILABLE:
+            raise RuntimeError("websockets package not installed")
+        
+        if not self.config.api_key:
+            raise ValueError("OpenAI API key not provided")
+        
+        self.state = RealtimeState.CONNECTING
+        
+        # Build URL
+        if self.config.endpoint:
+            # Azure or custom endpoint
+            url = f"{self.config.endpoint}/openai/realtime?api-version=2024-10-01-preview&deployment={self.config.model}"
+        else:
+            # OpenAI endpoint
+            url = f"wss://api.openai.com/v1/realtime?model={self.config.model}"
+        
+        # Build headers
+        headers = {}
+        if self.config.endpoint:
+            headers["api-key"] = self.config.api_key
+        else:
+            headers["Authorization"] = f"Bearer {self.config.api_key}"
+            headers["OpenAI-Beta"] = "realtime=v1"
+        
+        try:
+            logger.info(f"Connecting to Realtime API: {url}")
+            self._ws = await websockets.connect(url, extra_headers=headers)
+            
+            # Send session configuration
+            await self._configure_session()
+            
+            # Start receive loop
+            self._receive_task = asyncio.create_task(self._receive_loop())
+            
+            self.state = RealtimeState.CONNECTED
+            logger.info("Realtime API connected successfully")
+            
+        except Exception as e:
+            self.state = RealtimeState.ERROR
+            logger.error(f"Realtime API connection failed: {e}")
+            raise
+    
+    async def _configure_session(self) -> None:
+        """Send session configuration to server."""
+        session_config = {
+            "type": "session.update",
+            "session": {
+                "modalities": ["text", "audio"],
+                "instructions": self.config.instructions,
+                "voice": self.config.voice,
+                "input_audio_format": self.config.input_audio_format,
+                "output_audio_format": self.config.output_audio_format,
+                "turn_detection": self.config.turn_detection,
+            }
+        }
+        
+        if self.config.tools:
+            session_config["session"]["tools"] = self.config.tools
+        
+        await self._send(session_config)
+        logger.debug("Session configuration sent")
+    
+    async def _send(self, data: Dict[str, Any]) -> None:
+        """Send JSON data to server."""
+        if self._ws:
+            await self._ws.send(json.dumps(data))
+    
+    async def send_audio(self, audio_bytes: bytes) -> None:
+        """
+        Send audio to the Realtime API.
+        
+        Args:
+            audio_bytes: PCM audio data (16-bit, mono, 24kHz by default)
+        """
+        if self.state != RealtimeState.CONNECTED:
+            return
+        
+        # Encode audio as base64
+        audio_b64 = base64.standard_b64encode(audio_bytes).decode()
+        
+        await self._send({
+            "type": "input_audio_buffer.append",
+            "audio": audio_b64
+        })
+    
+    async def send_text(self, text: str) -> None:
+        """
+        Send text input (bypassing audio).
+        
+        Args:
+            text: User text input
+        """
+        if self.state != RealtimeState.CONNECTED:
+            return
+        
+        # Create a conversation item with user text
+        await self._send({
+            "type": "conversation.item.create",
+            "item": {
+                "type": "message",
+                "role": "user",
+                "content": [{"type": "input_text", "text": text}]
+            }
+        })
+        
+        # Trigger response
+        await self._send({"type": "response.create"})
+    
+    async def cancel_response(self) -> None:
+        """Cancel the current response (for barge-in)."""
+        if self.state != RealtimeState.CONNECTED:
+            return
+        
+        await self._send({"type": "response.cancel"})
+        logger.debug("Response cancelled")
+    
+    async def commit_audio(self) -> None:
+        """Commit the audio buffer and trigger response."""
+        if self.state != RealtimeState.CONNECTED:
+            return
+        
+        await self._send({"type": "input_audio_buffer.commit"})
+        await self._send({"type": "response.create"})
+    
+    async def clear_audio_buffer(self) -> None:
+        """Clear the input audio buffer."""
+        if self.state != RealtimeState.CONNECTED:
+            return
+        
+        await self._send({"type": "input_audio_buffer.clear"})
+    
+    async def submit_function_result(self, call_id: str, result: str) -> None:
+        """
+        Submit function call result.
+        
+        Args:
+            call_id: The function call ID
+            result: JSON string result
+        """
+        if self.state != RealtimeState.CONNECTED:
+            return
+        
+        await self._send({
+            "type": "conversation.item.create",
+            "item": {
+                "type": "function_call_output",
+                "call_id": call_id,
+                "output": result
+            }
+        })
+        
+        # Trigger response with the function result
+        await self._send({"type": "response.create"})
+    
+    async def _receive_loop(self) -> None:
+        """Receive and process messages from the Realtime API."""
+        if not self._ws:
+            return
+        
+        try:
+            async for message in self._ws:
+                try:
+                    data = json.loads(message)
+                    await self._handle_event(data)
+                except json.JSONDecodeError:
+                    logger.warning(f"Invalid JSON received: {message[:100]}")
+        
+        except asyncio.CancelledError:
+            logger.debug("Receive loop cancelled")
+        except websockets.ConnectionClosed as e:
+            logger.info(f"WebSocket closed: {e}")
+            self.state = RealtimeState.DISCONNECTED
+        except Exception as e:
+            logger.error(f"Receive loop error: {e}")
+            self.state = RealtimeState.ERROR
+    
+    async def _handle_event(self, data: Dict[str, Any]) -> None:
+        """Handle incoming event from Realtime API."""
+        event_type = data.get("type", "unknown")
+        
+        # Audio delta - streaming audio output
+        if event_type == "response.audio.delta":
+            if "delta" in data:
+                audio_bytes = base64.standard_b64decode(data["delta"])
+                await self._emit("on_audio", audio_bytes)
+        
+        # Audio transcript delta - streaming text
+        elif event_type == "response.audio_transcript.delta":
+            if "delta" in data:
+                await self._emit("on_transcript", data["delta"], "assistant", False)
+        
+        # Audio transcript done
+        elif event_type == "response.audio_transcript.done":
+            if "transcript" in data:
+                await self._emit("on_transcript", data["transcript"], "assistant", True)
+        
+        # Input audio transcript (user speech)
+        elif event_type == "conversation.item.input_audio_transcription.completed":
+            if "transcript" in data:
+                await self._emit("on_transcript", data["transcript"], "user", True)
+        
+        # Speech started (server VAD detected speech)
+        elif event_type == "input_audio_buffer.speech_started":
+            await self._emit("on_speech_started", data.get("audio_start_ms", 0))
+        
+        # Speech stopped
+        elif event_type == "input_audio_buffer.speech_stopped":
+            await self._emit("on_speech_stopped", data.get("audio_end_ms", 0))
+        
+        # Response started
+        elif event_type == "response.created":
+            await self._emit("on_response_started", data.get("response", {}))
+        
+        # Response done
+        elif event_type == "response.done":
+            await self._emit("on_response_done", data.get("response", {}))
+        
+        # Function call
+        elif event_type == "response.function_call_arguments.done":
+            call_id = data.get("call_id")
+            name = data.get("name")
+            arguments = data.get("arguments", "{}")
+            await self._emit("on_function_call", call_id, name, arguments)
+        
+        # Error
+        elif event_type == "error":
+            error = data.get("error", {})
+            logger.error(f"Realtime API error: {error}")
+            await self._emit("on_error", error)
+        
+        # Session events
+        elif event_type == "session.created":
+            logger.info("Session created")
+        elif event_type == "session.updated":
+            logger.debug("Session updated")
+        
+        else:
+            logger.debug(f"Unhandled event type: {event_type}")
+    
+    async def disconnect(self) -> None:
+        """Disconnect from Realtime API."""
+        self._cancel_event.set()
+        
+        if self._receive_task:
+            self._receive_task.cancel()
+            try:
+                await self._receive_task
+            except asyncio.CancelledError:
+                pass
+        
+        if self._ws:
+            await self._ws.close()
+            self._ws = None
+        
+        self.state = RealtimeState.DISCONNECTED
+        logger.info("Realtime API disconnected")
+
+
+class RealtimePipeline:
+    """
+    Pipeline adapter for RealtimeService.
+    
+    Provides a compatible interface with DuplexPipeline but uses
+    OpenAI Realtime API for all processing.
+    """
+    
+    def __init__(
+        self,
+        transport,
+        session_id: str,
+        config: Optional[RealtimeConfig] = None
+    ):
+        """
+        Initialize Realtime pipeline.
+        
+        Args:
+            transport: Transport for sending audio/events
+            session_id: Session identifier
+            config: Realtime configuration
+        """
+        self.transport = transport
+        self.session_id = session_id
+        
+        self.service = RealtimeService(config)
+        
+        # Register callbacks
+        self.service.on("on_audio", self._on_audio)
+        self.service.on("on_transcript", self._on_transcript)
+        self.service.on("on_speech_started", self._on_speech_started)
+        self.service.on("on_speech_stopped", self._on_speech_stopped)
+        self.service.on("on_response_started", self._on_response_started)
+        self.service.on("on_response_done", self._on_response_done)
+        self.service.on("on_error", self._on_error)
+        
+        self._is_speaking = False
+        self._running = True
+        
+        logger.info(f"RealtimePipeline initialized for session {session_id}")
+    
+    async def start(self) -> None:
+        """Start the pipeline."""
+        await self.service.connect()
+    
+    async def process_audio(self, pcm_bytes: bytes) -> None:
+        """
+        Process incoming audio.
+        
+        Note: Realtime API expects 24kHz audio by default.
+        You may need to resample from 16kHz.
+        """
+        if not self._running:
+            return
+        
+        # TODO: Resample from 16kHz to 24kHz if needed
+        await self.service.send_audio(pcm_bytes)
+    
+    async def process_text(self, text: str) -> None:
+        """Process text input."""
+        if not self._running:
+            return
+        
+        await self.service.send_text(text)
+    
+    async def interrupt(self) -> None:
+        """Interrupt current response."""
+        await self.service.cancel_response()
+        await self.transport.send_event({
+            "event": "interrupt",
+            "trackId": self.session_id,
+            "timestamp": self._get_timestamp_ms()
+        })
+    
+    async def cleanup(self) -> None:
+        """Cleanup resources."""
+        self._running = False
+        await self.service.disconnect()
+    
+    # Event handlers
+    
+    async def _on_audio(self, audio_bytes: bytes) -> None:
+        """Handle audio output."""
+        await self.transport.send_audio(audio_bytes)
+    
+    async def _on_transcript(self, text: str, role: str, is_final: bool) -> None:
+        """Handle transcript."""
+        logger.info(f"[{role.upper()}] {text[:50]}..." if len(text) > 50 else f"[{role.upper()}] {text}")
+    
+    async def _on_speech_started(self, start_ms: int) -> None:
+        """Handle user speech start."""
+        self._is_speaking = True
+        await self.transport.send_event({
+            "event": "speaking",
+            "trackId": self.session_id,
+            "timestamp": self._get_timestamp_ms(),
+            "startTime": start_ms
+        })
+        
+        # Cancel any ongoing response (barge-in)
+        await self.service.cancel_response()
+    
+    async def _on_speech_stopped(self, end_ms: int) -> None:
+        """Handle user speech stop."""
+        self._is_speaking = False
+        await self.transport.send_event({
+            "event": "silence",
+            "trackId": self.session_id,
+            "timestamp": self._get_timestamp_ms(),
+            "duration": end_ms
+        })
+    
+    async def _on_response_started(self, response: Dict) -> None:
+        """Handle response start."""
+        await self.transport.send_event({
+            "event": "trackStart",
+            "trackId": self.session_id,
+            "timestamp": self._get_timestamp_ms()
+        })
+    
+    async def _on_response_done(self, response: Dict) -> None:
+        """Handle response complete."""
+        await self.transport.send_event({
+            "event": "trackEnd",
+            "trackId": self.session_id,
+            "timestamp": self._get_timestamp_ms()
+        })
+    
+    async def _on_error(self, error: Dict) -> None:
+        """Handle error."""
+        await self.transport.send_event({
+            "event": "error",
+            "trackId": self.session_id,
+            "timestamp": self._get_timestamp_ms(),
+            "sender": "realtime",
+            "error": str(error)
+        })
+    
+    def _get_timestamp_ms(self) -> int:
+        """Get current timestamp in milliseconds."""
+        import time
+        return int(time.time() * 1000)
+    
+    @property
+    def is_speaking(self) -> bool:
+        """Check if user is speaking."""
+        return self._is_speaking

services/siliconflow_tts.py

@@ -0,0 +1,255 @@
+"""SiliconFlow TTS Service with streaming support.
+
+Uses SiliconFlow's CosyVoice2 or MOSS-TTSD models for low-latency
+text-to-speech synthesis with streaming.
+
+API Docs: https://docs.siliconflow.cn/cn/api-reference/audio/create-speech
+"""
+
+import os
+import asyncio
+import aiohttp
+from typing import AsyncIterator, Optional
+from loguru import logger
+
+from services.base import BaseTTSService, TTSChunk, ServiceState
+
+
+class SiliconFlowTTSService(BaseTTSService):
+    """
+    SiliconFlow TTS service with streaming support.
+    
+    Supports CosyVoice2-0.5B and MOSS-TTSD-v0.5 models.
+    """
+    
+    # Available voices
+    VOICES = {
+        "alex": "FunAudioLLM/CosyVoice2-0.5B:alex",
+        "anna": "FunAudioLLM/CosyVoice2-0.5B:anna",
+        "bella": "FunAudioLLM/CosyVoice2-0.5B:bella",
+        "benjamin": "FunAudioLLM/CosyVoice2-0.5B:benjamin",
+        "charles": "FunAudioLLM/CosyVoice2-0.5B:charles",
+        "claire": "FunAudioLLM/CosyVoice2-0.5B:claire",
+        "david": "FunAudioLLM/CosyVoice2-0.5B:david",
+        "diana": "FunAudioLLM/CosyVoice2-0.5B:diana",
+    }
+    
+    def __init__(
+        self,
+        api_key: Optional[str] = None,
+        voice: str = "anna",
+        model: str = "FunAudioLLM/CosyVoice2-0.5B",
+        sample_rate: int = 16000,
+        speed: float = 1.0
+    ):
+        """
+        Initialize SiliconFlow TTS service.
+        
+        Args:
+            api_key: SiliconFlow API key (defaults to SILICONFLOW_API_KEY env var)
+            voice: Voice name (alex, anna, bella, benjamin, charles, claire, david, diana)
+            model: Model name
+            sample_rate: Output sample rate (8000, 16000, 24000, 32000, 44100)
+            speed: Speech speed (0.25 to 4.0)
+        """
+        # Resolve voice name
+        if voice in self.VOICES:
+            full_voice = self.VOICES[voice]
+        else:
+            full_voice = voice
+            
+        super().__init__(voice=full_voice, sample_rate=sample_rate, speed=speed)
+        
+        self.api_key = api_key or os.getenv("SILICONFLOW_API_KEY")
+        self.model = model
+        self.api_url = "https://api.siliconflow.cn/v1/audio/speech"
+        
+        self._session: Optional[aiohttp.ClientSession] = None
+        self._cancel_event = asyncio.Event()
+    
+    async def connect(self) -> None:
+        """Initialize HTTP session."""
+        if not self.api_key:
+            raise ValueError("SiliconFlow API key not provided. Set SILICONFLOW_API_KEY env var.")
+        
+        self._session = aiohttp.ClientSession(
+            headers={
+                "Authorization": f"Bearer {self.api_key}",
+                "Content-Type": "application/json"
+            }
+        )
+        self.state = ServiceState.CONNECTED
+        logger.info(f"SiliconFlow TTS service ready: voice={self.voice}, model={self.model}")
+    
+    async def disconnect(self) -> None:
+        """Close HTTP session."""
+        if self._session:
+            await self._session.close()
+            self._session = None
+        self.state = ServiceState.DISCONNECTED
+        logger.info("SiliconFlow TTS service disconnected")
+    
+    async def synthesize(self, text: str) -> bytes:
+        """Synthesize complete audio for text."""
+        audio_data = b""
+        async for chunk in self.synthesize_stream(text):
+            audio_data += chunk.audio
+        return audio_data
+    
+    async def synthesize_stream(self, text: str) -> AsyncIterator[TTSChunk]:
+        """
+        Synthesize audio in streaming mode.
+        
+        Args:
+            text: Text to synthesize
+            
+        Yields:
+            TTSChunk objects with PCM audio
+        """
+        if not self._session:
+            raise RuntimeError("TTS service not connected")
+        
+        if not text.strip():
+            return
+        
+        self._cancel_event.clear()
+        
+        payload = {
+            "model": self.model,
+            "input": text,
+            "voice": self.voice,
+            "response_format": "pcm",
+            "sample_rate": self.sample_rate,
+            "stream": True,
+            "speed": self.speed
+        }
+        
+        try:
+            async with self._session.post(self.api_url, json=payload) as response:
+                if response.status != 200:
+                    error_text = await response.text()
+                    logger.error(f"SiliconFlow TTS error: {response.status} - {error_text}")
+                    return
+                
+                # Stream audio chunks
+                chunk_size = self.sample_rate * 2 // 10  # 100ms chunks
+                buffer = b""
+                
+                async for chunk in response.content.iter_any():
+                    if self._cancel_event.is_set():
+                        logger.info("TTS synthesis cancelled")
+                        return
+                    
+                    buffer += chunk
+                    
+                    # Yield complete chunks
+                    while len(buffer) >= chunk_size:
+                        audio_chunk = buffer[:chunk_size]
+                        buffer = buffer[chunk_size:]
+                        
+                        yield TTSChunk(
+                            audio=audio_chunk,
+                            sample_rate=self.sample_rate,
+                            is_final=False
+                        )
+                
+                # Yield remaining buffer
+                if buffer:
+                    yield TTSChunk(
+                        audio=buffer,
+                        sample_rate=self.sample_rate,
+                        is_final=True
+                    )
+                    
+        except asyncio.CancelledError:
+            logger.info("TTS synthesis cancelled via asyncio")
+            raise
+        except Exception as e:
+            logger.error(f"TTS synthesis error: {e}")
+            raise
+    
+    async def cancel(self) -> None:
+        """Cancel ongoing synthesis."""
+        self._cancel_event.set()
+
+
+class StreamingTTSAdapter:
+    """
+    Adapter for streaming LLM text to TTS with sentence-level chunking.
+    
+    This reduces latency by starting TTS as soon as a complete sentence
+    is received from the LLM, rather than waiting for the full response.
+    """
+    
+    # Sentence delimiters
+    SENTENCE_ENDS = {'.', '!', '?', '。', '！', '？', '；', '\n'}
+    
+    def __init__(self, tts_service: BaseTTSService, transport, session_id: str):
+        self.tts_service = tts_service
+        self.transport = transport
+        self.session_id = session_id
+        self._buffer = ""
+        self._cancel_event = asyncio.Event()
+        self._is_speaking = False
+    
+    async def process_text_chunk(self, text_chunk: str) -> None:
+        """
+        Process a text chunk from LLM and trigger TTS when sentence is complete.
+        
+        Args:
+            text_chunk: Text chunk from LLM streaming
+        """
+        if self._cancel_event.is_set():
+            return
+        
+        self._buffer += text_chunk
+        
+        # Check for sentence completion
+        for i, char in enumerate(self._buffer):
+            if char in self.SENTENCE_ENDS:
+                # Found sentence end, synthesize up to this point
+                sentence = self._buffer[:i+1].strip()
+                self._buffer = self._buffer[i+1:]
+                
+                if sentence:
+                    await self._speak_sentence(sentence)
+                break
+    
+    async def flush(self) -> None:
+        """Flush remaining buffer."""
+        if self._buffer.strip() and not self._cancel_event.is_set():
+            await self._speak_sentence(self._buffer.strip())
+        self._buffer = ""
+    
+    async def _speak_sentence(self, text: str) -> None:
+        """Synthesize and send a sentence."""
+        if not text or self._cancel_event.is_set():
+            return
+        
+        self._is_speaking = True
+        
+        try:
+            async for chunk in self.tts_service.synthesize_stream(text):
+                if self._cancel_event.is_set():
+                    break
+                await self.transport.send_audio(chunk.audio)
+                await asyncio.sleep(0.01)  # Prevent flooding
+        except Exception as e:
+            logger.error(f"TTS speak error: {e}")
+        finally:
+            self._is_speaking = False
+    
+    def cancel(self) -> None:
+        """Cancel ongoing speech."""
+        self._cancel_event.set()
+        self._buffer = ""
+    
+    def reset(self) -> None:
+        """Reset for new turn."""
+        self._cancel_event.clear()
+        self._buffer = ""
+        self._is_speaking = False
+    
+    @property
+    def is_speaking(self) -> bool:
+        return self._is_speaking

services/tts.py

@@ -0,0 +1,271 @@
+"""TTS (Text-to-Speech) Service implementations.
+
+Provides multiple TTS backend options including edge-tts (free)
+and placeholder for cloud services.
+"""
+
+import os
+import io
+import asyncio
+import struct
+from typing import AsyncIterator, Optional
+from loguru import logger
+
+from services.base import BaseTTSService, TTSChunk, ServiceState
+
+# Try to import edge-tts
+try:
+    import edge_tts
+    EDGE_TTS_AVAILABLE = True
+except ImportError:
+    EDGE_TTS_AVAILABLE = False
+    logger.warning("edge-tts not available - EdgeTTS service will be disabled")
+
+
+class EdgeTTSService(BaseTTSService):
+    """
+    Microsoft Edge TTS service.
+    
+    Uses edge-tts library for free, high-quality speech synthesis.
+    Supports streaming for low-latency playback.
+    """
+    
+    # Voice mapping for common languages
+    VOICE_MAP = {
+        "en": "en-US-JennyNeural",
+        "en-US": "en-US-JennyNeural",
+        "en-GB": "en-GB-SoniaNeural",
+        "zh": "zh-CN-XiaoxiaoNeural",
+        "zh-CN": "zh-CN-XiaoxiaoNeural",
+        "zh-TW": "zh-TW-HsiaoChenNeural",
+        "ja": "ja-JP-NanamiNeural",
+        "ko": "ko-KR-SunHiNeural",
+        "fr": "fr-FR-DeniseNeural",
+        "de": "de-DE-KatjaNeural",
+        "es": "es-ES-ElviraNeural",
+    }
+    
+    def __init__(
+        self,
+        voice: str = "en-US-JennyNeural",
+        sample_rate: int = 16000,
+        speed: float = 1.0
+    ):
+        """
+        Initialize Edge TTS service.
+        
+        Args:
+            voice: Voice name (e.g., "en-US-JennyNeural") or language code (e.g., "en")
+            sample_rate: Target sample rate (will be resampled)
+            speed: Speech speed multiplier
+        """
+        # Resolve voice from language code if needed
+        if voice in self.VOICE_MAP:
+            voice = self.VOICE_MAP[voice]
+        
+        super().__init__(voice=voice, sample_rate=sample_rate, speed=speed)
+        self._cancel_event = asyncio.Event()
+    
+    async def connect(self) -> None:
+        """Edge TTS doesn't require explicit connection."""
+        if not EDGE_TTS_AVAILABLE:
+            raise RuntimeError("edge-tts package not installed")
+        self.state = ServiceState.CONNECTED
+        logger.info(f"Edge TTS service ready: voice={self.voice}")
+    
+    async def disconnect(self) -> None:
+        """Edge TTS doesn't require explicit disconnection."""
+        self.state = ServiceState.DISCONNECTED
+        logger.info("Edge TTS service disconnected")
+    
+    def _get_rate_string(self) -> str:
+        """Convert speed to rate string for edge-tts."""
+        # edge-tts uses percentage format: "+0%", "-10%", "+20%"
+        percentage = int((self.speed - 1.0) * 100)
+        if percentage >= 0:
+            return f"+{percentage}%"
+        return f"{percentage}%"
+    
+    async def synthesize(self, text: str) -> bytes:
+        """
+        Synthesize complete audio for text.
+        
+        Args:
+            text: Text to synthesize
+            
+        Returns:
+            PCM audio data (16-bit, mono, 16kHz)
+        """
+        if not EDGE_TTS_AVAILABLE:
+            raise RuntimeError("edge-tts not available")
+        
+        # Collect all chunks
+        audio_data = b""
+        async for chunk in self.synthesize_stream(text):
+            audio_data += chunk.audio
+        
+        return audio_data
+    
+    async def synthesize_stream(self, text: str) -> AsyncIterator[TTSChunk]:
+        """
+        Synthesize audio in streaming mode.
+        
+        Args:
+            text: Text to synthesize
+            
+        Yields:
+            TTSChunk objects with PCM audio
+        """
+        if not EDGE_TTS_AVAILABLE:
+            raise RuntimeError("edge-tts not available")
+        
+        self._cancel_event.clear()
+        
+        try:
+            communicate = edge_tts.Communicate(
+                text,
+                voice=self.voice,
+                rate=self._get_rate_string()
+            )
+            
+            # edge-tts outputs MP3, we need to decode to PCM
+            # For now, collect MP3 chunks and yield after conversion
+            mp3_data = b""
+            
+            async for chunk in communicate.stream():
+                # Check for cancellation
+                if self._cancel_event.is_set():
+                    logger.info("TTS synthesis cancelled")
+                    return
+                
+                if chunk["type"] == "audio":
+                    mp3_data += chunk["data"]
+            
+            # Convert MP3 to PCM
+            if mp3_data:
+                pcm_data = await self._convert_mp3_to_pcm(mp3_data)
+                if pcm_data:
+                    # Yield in chunks for streaming playback
+                    chunk_size = self.sample_rate * 2 // 10  # 100ms chunks
+                    for i in range(0, len(pcm_data), chunk_size):
+                        if self._cancel_event.is_set():
+                            return
+                        
+                        chunk_data = pcm_data[i:i + chunk_size]
+                        yield TTSChunk(
+                            audio=chunk_data,
+                            sample_rate=self.sample_rate,
+                            is_final=(i + chunk_size >= len(pcm_data))
+                        )
+        
+        except asyncio.CancelledError:
+            logger.info("TTS synthesis cancelled via asyncio")
+            raise
+        except Exception as e:
+            logger.error(f"TTS synthesis error: {e}")
+            raise
+    
+    async def _convert_mp3_to_pcm(self, mp3_data: bytes) -> bytes:
+        """
+        Convert MP3 audio to PCM.
+        
+        Uses pydub or ffmpeg for conversion.
+        """
+        try:
+            # Try using pydub (requires ffmpeg)
+            from pydub import AudioSegment
+            
+            # Load MP3 from bytes
+            audio = AudioSegment.from_mp3(io.BytesIO(mp3_data))
+            
+            # Convert to target format
+            audio = audio.set_frame_rate(self.sample_rate)
+            audio = audio.set_channels(1)
+            audio = audio.set_sample_width(2)  # 16-bit
+            
+            # Export as raw PCM
+            return audio.raw_data
+            
+        except ImportError:
+            logger.warning("pydub not available, trying fallback")
+            # Fallback: Use subprocess to call ffmpeg directly
+            return await self._ffmpeg_convert(mp3_data)
+        except Exception as e:
+            logger.error(f"Audio conversion error: {e}")
+            return b""
+    
+    async def _ffmpeg_convert(self, mp3_data: bytes) -> bytes:
+        """Convert MP3 to PCM using ffmpeg subprocess."""
+        try:
+            process = await asyncio.create_subprocess_exec(
+                "ffmpeg",
+                "-i", "pipe:0",
+                "-f", "s16le",
+                "-acodec", "pcm_s16le",
+                "-ar", str(self.sample_rate),
+                "-ac", "1",
+                "pipe:1",
+                stdin=asyncio.subprocess.PIPE,
+                stdout=asyncio.subprocess.PIPE,
+                stderr=asyncio.subprocess.DEVNULL
+            )
+            
+            stdout, _ = await process.communicate(input=mp3_data)
+            return stdout
+            
+        except Exception as e:
+            logger.error(f"ffmpeg conversion error: {e}")
+            return b""
+    
+    async def cancel(self) -> None:
+        """Cancel ongoing synthesis."""
+        self._cancel_event.set()
+
+
+class MockTTSService(BaseTTSService):
+    """
+    Mock TTS service for testing without actual synthesis.
+    
+    Generates silence or simple tones.
+    """
+    
+    def __init__(
+        self,
+        voice: str = "mock",
+        sample_rate: int = 16000,
+        speed: float = 1.0
+    ):
+        super().__init__(voice=voice, sample_rate=sample_rate, speed=speed)
+    
+    async def connect(self) -> None:
+        self.state = ServiceState.CONNECTED
+        logger.info("Mock TTS service connected")
+    
+    async def disconnect(self) -> None:
+        self.state = ServiceState.DISCONNECTED
+        logger.info("Mock TTS service disconnected")
+    
+    async def synthesize(self, text: str) -> bytes:
+        """Generate silence based on text length."""
+        # Approximate: 100ms per word
+        word_count = len(text.split())
+        duration_ms = word_count * 100
+        samples = int(self.sample_rate * duration_ms / 1000)
+        
+        # Generate silence (zeros)
+        return bytes(samples * 2)  # 16-bit = 2 bytes per sample
+    
+    async def synthesize_stream(self, text: str) -> AsyncIterator[TTSChunk]:
+        """Generate silence chunks."""
+        audio = await self.synthesize(text)
+        
+        # Yield in 100ms chunks
+        chunk_size = self.sample_rate * 2 // 10
+        for i in range(0, len(audio), chunk_size):
+            chunk_data = audio[i:i + chunk_size]
+            yield TTSChunk(
+                audio=chunk_data,
+                sample_rate=self.sample_rate,
+                is_final=(i + chunk_size >= len(audio))
+            )
+            await asyncio.sleep(0.05)  # Simulate processing time