fix long run bug

examples/mic_client.py

@@ -17,6 +17,7 @@ import argparse
 import asyncio
 import json
 import sys
+import time
 import threading
 import queue
 from pathlib import Path
@@ -92,6 +93,14 @@ class MicrophoneClient:
         # State
         self.is_recording = True
         self.is_playing = True
+        
+        # TTFB tracking (Time to First Byte)
+        self.request_start_time = None
+        self.first_audio_received = False
+        
+        # Interrupt handling - discard audio until next trackStart
+        self._discard_audio = False
+        self._audio_sequence = 0  # Track audio sequence to detect stale chunks
     
     async def connect(self) -> None:
         """Connect to WebSocket server."""
@@ -117,6 +126,10 @@ class MicrophoneClient:
     
     async def send_chat(self, text: str) -> None:
         """Send chat message (text input)."""
+        # Reset TTFB tracking for new request
+        self.request_start_time = time.time()
+        self.first_audio_received = False
+        
         await self.send_command({
             "command": "chat",
             "text": text
@@ -236,9 +249,21 @@ class MicrophoneClient:
                         # Audio data received
                         self.bytes_received += len(message)
                         
+                        # Check if we should discard this audio (after interrupt)
+                        if self._discard_audio:
+                            duration_ms = len(message) / (self.sample_rate * 2) * 1000
+                            print(f"← Audio: {duration_ms:.0f}ms (DISCARDED - waiting for new track)")
+                            continue
+                        
                         if self.is_playing:
                             self._add_audio_to_buffer(message)
                         
+                        # Calculate and display TTFB for first audio packet
+                        if not self.first_audio_received and self.request_start_time:
+                            client_ttfb_ms = (time.time() - self.request_start_time) * 1000
+                            self.first_audio_received = True
+                            print(f"← [TTFB] Client first audio latency: {client_ttfb_ms:.0f}ms")
+                        
                         # Show progress (less verbose)
                         with self.audio_output_lock:
                             buffer_ms = len(self.audio_output_buffer) / (self.sample_rate * 2) * 1000
@@ -285,20 +310,36 @@ class MicrophoneClient:
                 # Interim result - show with indicator (overwrite same line)
                 display_text = text[:60] + "..." if len(text) > 60 else text
                 print(f"  [listening] {display_text}".ljust(80), end="\r")
+        elif event_type == "ttfb":
+            # Server-side TTFB event
+            latency_ms = event.get("latencyMs", 0)
+            print(f"← [TTFB] Server reported latency: {latency_ms}ms")
         elif event_type == "trackStart":
             print("← Bot started speaking")
+            # IMPORTANT: Accept audio again after trackStart
+            self._discard_audio = False
+            self._audio_sequence += 1
+            # Reset TTFB tracking for voice responses (when no chat was sent)
+            if self.request_start_time is None:
+                self.request_start_time = time.time()
+                self.first_audio_received = False
             # Clear any old audio in buffer
             with self.audio_output_lock:
                 self.audio_output_buffer = b""
         elif event_type == "trackEnd":
             print("← Bot finished speaking")
+            # Reset TTFB tracking after response completes
+            self.request_start_time = None
+            self.first_audio_received = False
         elif event_type == "interrupt":
             print("← Bot interrupted!")
-            # IMPORTANT: Clear audio buffer immediately on interrupt
+            # IMPORTANT: Discard all audio until next trackStart
+            self._discard_audio = True
+            # Clear audio buffer immediately
             with self.audio_output_lock:
                 buffer_ms = len(self.audio_output_buffer) / (self.sample_rate * 2) * 1000
                 self.audio_output_buffer = b""
-                print(f"   (cleared {buffer_ms:.0f}ms of buffered audio)")
+                print(f"   (cleared {buffer_ms:.0f}ms, discarding audio until new track)")
         elif event_type == "error":
             print(f"← Error: {event.get('error')}")
         elif event_type == "hangup":