Merge pull request #795 from pipecat-ai/vp-nvidia-riva
[WIP] add nvidia riva
This commit is contained in:
Vanessa Pyne
@@ -1,5 +1,5 @@
|
||||
build~=1.2.1
|
||||
grpcio-tools~=1.62.2
|
||||
grpcio-tools~=1.65.4
|
||||
pip-tools~=7.4.1
|
||||
pyright~=1.1.376
|
||||
pytest~=8.3.2
|
||||
|
||||
56
examples/foundational/01c-fastpitch.py
Normal file
56
examples/foundational/01c-fastpitch.py
Normal file
@@ -0,0 +1,56 @@
|
||||
#
|
||||
# Copyright (c) 2024, Daily
|
||||
#
|
||||
# SPDX-License-Identifier: BSD 2-Clause License
|
||||
#
|
||||
|
||||
import asyncio
|
||||
import aiohttp
|
||||
import os
|
||||
import sys
|
||||
|
||||
from pipecat.frames.frames import EndFrame, TTSSpeakFrame
|
||||
from pipecat.pipeline.pipeline import Pipeline
|
||||
from pipecat.pipeline.task import PipelineTask
|
||||
from pipecat.pipeline.runner import PipelineRunner
|
||||
from pipecat.services.riva import FastpitchTTSService
|
||||
from pipecat.transports.services.daily import DailyParams, DailyTransport
|
||||
|
||||
from runner import configure
|
||||
|
||||
from loguru import logger
|
||||
|
||||
from dotenv import load_dotenv
|
||||
|
||||
load_dotenv(override=True)
|
||||
|
||||
logger.remove(0)
|
||||
logger.add(sys.stderr, level="DEBUG")
|
||||
|
||||
|
||||
async def main():
|
||||
async with aiohttp.ClientSession() as session:
|
||||
(room_url, _) = await configure(session)
|
||||
|
||||
transport = DailyTransport(
|
||||
room_url, None, "Say One Thing", DailyParams(audio_out_enabled=True)
|
||||
)
|
||||
|
||||
tts = FastpitchTTSService(api_key=os.getenv("NVIDIA_API_KEY"))
|
||||
|
||||
runner = PipelineRunner()
|
||||
|
||||
task = PipelineTask(Pipeline([tts, transport.output()]))
|
||||
|
||||
# Register an event handler so we can play the audio when the
|
||||
# participant joins.
|
||||
@transport.event_handler("on_first_participant_joined")
|
||||
async def on_first_participant_joined(transport, participant):
|
||||
participant_name = participant.get("info", {}).get("userName", "")
|
||||
await task.queue_frames([TTSSpeakFrame(f"Aloha, {participant_name}!"), EndFrame()])
|
||||
|
||||
await runner.run(task)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
asyncio.run(main())
|
||||
92
examples/foundational/07r-interruptible-riva-nim.py
Normal file
92
examples/foundational/07r-interruptible-riva-nim.py
Normal file
@@ -0,0 +1,92 @@
|
||||
#
|
||||
# Copyright (c) 2024, Daily
|
||||
#
|
||||
# SPDX-License-Identifier: BSD 2-Clause License
|
||||
#
|
||||
|
||||
import asyncio
|
||||
import os
|
||||
import sys
|
||||
|
||||
import aiohttp
|
||||
from dotenv import load_dotenv
|
||||
from loguru import logger
|
||||
from runner import configure
|
||||
|
||||
from pipecat.audio.vad.silero import SileroVADAnalyzer
|
||||
from pipecat.frames.frames import LLMMessagesFrame
|
||||
from pipecat.pipeline.pipeline import Pipeline
|
||||
from pipecat.pipeline.runner import PipelineRunner
|
||||
from pipecat.pipeline.task import PipelineParams, PipelineTask
|
||||
from pipecat.processors.aggregators.openai_llm_context import OpenAILLMContext
|
||||
from pipecat.services.nim import NimLLMService
|
||||
from pipecat.services.riva import FastpitchTTSService, ParakeetSTTService
|
||||
from pipecat.transports.services.daily import DailyParams, DailyTransport
|
||||
|
||||
load_dotenv(override=True)
|
||||
|
||||
logger.remove(0)
|
||||
logger.add(sys.stderr, level="DEBUG")
|
||||
|
||||
|
||||
async def main():
|
||||
async with aiohttp.ClientSession() as session:
|
||||
(room_url, _) = await configure(session)
|
||||
|
||||
transport = DailyTransport(
|
||||
room_url,
|
||||
None,
|
||||
"Respond bot",
|
||||
DailyParams(
|
||||
audio_out_enabled=True,
|
||||
vad_enabled=True,
|
||||
vad_analyzer=SileroVADAnalyzer(),
|
||||
vad_audio_passthrough=True,
|
||||
),
|
||||
)
|
||||
|
||||
stt = ParakeetSTTService(api_key=os.getenv("NVIDIA_API_KEY"))
|
||||
|
||||
llm = NimLLMService(
|
||||
api_key=os.getenv("NVIDIA_API_KEY"), model="meta/llama-3.1-405b-instruct"
|
||||
)
|
||||
|
||||
tts = FastpitchTTSService(api_key=os.getenv("NVIDIA_API_KEY"))
|
||||
|
||||
messages = [
|
||||
{
|
||||
"role": "system",
|
||||
"content": "You are a helpful LLM in a WebRTC call. Your goal is to demonstrate your capabilities in a succinct way. Your output will be converted to audio so don't include special characters in your answers. Respond to what the user said in a creative and helpful way.",
|
||||
},
|
||||
]
|
||||
|
||||
context = OpenAILLMContext(messages)
|
||||
context_aggregator = llm.create_context_aggregator(context)
|
||||
|
||||
pipeline = Pipeline(
|
||||
[
|
||||
transport.input(), # Transport user input
|
||||
stt, # STT
|
||||
context_aggregator.user(), # User responses
|
||||
llm, # LLM
|
||||
tts, # TTS
|
||||
transport.output(), # Transport bot output
|
||||
context_aggregator.assistant(), # Assistant spoken responses
|
||||
]
|
||||
)
|
||||
|
||||
task = PipelineTask(pipeline, PipelineParams(allow_interruptions=True))
|
||||
|
||||
@transport.event_handler("on_first_participant_joined")
|
||||
async def on_first_participant_joined(transport, participant):
|
||||
# Kick off the conversation.
|
||||
messages.append({"role": "system", "content": "Please introduce yourself to the user."})
|
||||
await task.queue_frames([LLMMessagesFrame(messages)])
|
||||
|
||||
runner = PipelineRunner()
|
||||
|
||||
await runner.run(task)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
asyncio.run(main())
|
||||
@@ -25,7 +25,7 @@ dependencies = [
|
||||
"Markdown~=3.7",
|
||||
"numpy~=1.26.4",
|
||||
"Pillow~=10.4.0",
|
||||
"protobuf~=4.25.4",
|
||||
"protobuf~=5.26.1",
|
||||
"pydantic~=2.8.2",
|
||||
"pyloudnorm~=0.1.1",
|
||||
"resampy~=0.4.3",
|
||||
@@ -48,7 +48,7 @@ elevenlabs = [ "websockets~=13.1" ]
|
||||
examples = [ "python-dotenv~=1.0.1", "flask~=3.0.3", "flask_cors~=4.0.1" ]
|
||||
fal = [ "fal-client~=0.4.1" ]
|
||||
gladia = [ "websockets~=13.1" ]
|
||||
google = [ "google-generativeai~=0.8.3", "google-cloud-texttospeech~=2.17.2" ]
|
||||
google = [ "google-generativeai~=0.8.3", "google-cloud-texttospeech~=2.21.1" ]
|
||||
grok = [ "openai~=1.50.2" ]
|
||||
groq = [ "openai~=1.50.2" ]
|
||||
gstreamer = [ "pygobject~=3.48.2" ]
|
||||
@@ -63,7 +63,8 @@ nim = [ "openai~=1.50.2" ]
|
||||
noisereduce = [ "noisereduce~=3.0.3" ]
|
||||
openai = [ "openai~=1.50.2", "websockets~=13.1", "python-deepcompare~=1.0.1" ]
|
||||
openpipe = [ "openpipe~=4.38.0" ]
|
||||
playht = [ "pyht~=0.1.4", "websockets~=13.1" ]
|
||||
playht = [ "pyht~=0.1.8", "websockets~=13.1" ]
|
||||
riva = [ "nvidia-riva-client~=2.17.0" ]
|
||||
silero = [ "onnxruntime~=1.19.2" ]
|
||||
soundfile = [ "soundfile~=0.12.1" ]
|
||||
together = [ "openai~=1.50.2" ]
|
||||
|
||||
266
src/pipecat/services/riva.py
Normal file
266
src/pipecat/services/riva.py
Normal file
@@ -0,0 +1,266 @@
|
||||
#
|
||||
# Copyright (c) 2024, Daily
|
||||
#
|
||||
# SPDX-License-Identifier: BSD 2-Clause License
|
||||
#
|
||||
|
||||
import asyncio
|
||||
from typing import AsyncGenerator, Optional
|
||||
|
||||
from loguru import logger
|
||||
from pydantic.main import BaseModel
|
||||
|
||||
from pipecat.frames.frames import (
|
||||
CancelFrame,
|
||||
EndFrame,
|
||||
Frame,
|
||||
InterimTranscriptionFrame,
|
||||
StartFrame,
|
||||
TranscriptionFrame,
|
||||
TTSAudioRawFrame,
|
||||
TTSStartedFrame,
|
||||
TTSStoppedFrame,
|
||||
)
|
||||
from pipecat.services.ai_services import STTService, TTSService
|
||||
from pipecat.transcriptions.language import Language
|
||||
from pipecat.utils.time import time_now_iso8601
|
||||
|
||||
try:
|
||||
import riva.client
|
||||
|
||||
except ModuleNotFoundError as e:
|
||||
logger.error(f"Exception: {e}")
|
||||
logger.error(
|
||||
"In order to use nvidia riva TTS or STT, you need to `pip install pipecat-ai[riva]`. Also, set `NVIDIA_API_KEY` environment variable."
|
||||
)
|
||||
raise Exception(f"Missing module: {e}")
|
||||
|
||||
|
||||
class FastpitchTTSService(TTSService):
|
||||
class InputParams(BaseModel):
|
||||
language: Optional[Language] = Language.EN_US
|
||||
quality: Optional[int] = 20
|
||||
|
||||
def __init__(
|
||||
self,
|
||||
*,
|
||||
api_key: str,
|
||||
server: str = "grpc.nvcf.nvidia.com:443",
|
||||
voice_id: str = "English-US.Female-1",
|
||||
sample_rate: int = 24000,
|
||||
function_id: str = "0149dedb-2be8-4195-b9a0-e57e0e14f972",
|
||||
params: InputParams = InputParams(),
|
||||
**kwargs,
|
||||
):
|
||||
super().__init__(sample_rate=sample_rate, **kwargs)
|
||||
self._api_key = api_key
|
||||
self._voice_id = voice_id
|
||||
self._sample_rate = sample_rate
|
||||
self._language_code = params.language
|
||||
self._quality = params.quality
|
||||
|
||||
self.set_model_name("fastpitch-hifigan-tts")
|
||||
self.set_voice(voice_id)
|
||||
|
||||
metadata = [
|
||||
["function-id", function_id],
|
||||
["authorization", f"Bearer {api_key}"],
|
||||
]
|
||||
auth = riva.client.Auth(None, True, server, metadata)
|
||||
|
||||
self._service = riva.client.SpeechSynthesisService(auth)
|
||||
|
||||
# warm up the service
|
||||
config_response = self._service.stub.GetRivaSynthesisConfig(
|
||||
riva.client.proto.riva_tts_pb2.RivaSynthesisConfigRequest()
|
||||
)
|
||||
|
||||
async def run_tts(self, text: str) -> AsyncGenerator[Frame, None]:
|
||||
def read_audio_responses():
|
||||
try:
|
||||
responses = self._service.synthesize_online(
|
||||
text,
|
||||
self._voice_id,
|
||||
self._language_code,
|
||||
sample_rate_hz=self._sample_rate,
|
||||
audio_prompt_file=None,
|
||||
quality=self._quality,
|
||||
custom_dictionary={},
|
||||
)
|
||||
return responses
|
||||
except Exception as e:
|
||||
logger.error(f"{self} exception: {e}")
|
||||
return []
|
||||
|
||||
await self.start_ttfb_metrics()
|
||||
yield TTSStartedFrame()
|
||||
|
||||
logger.debug(f"Generating TTS: [{text}]")
|
||||
responses = await asyncio.to_thread(read_audio_responses)
|
||||
|
||||
for resp in responses:
|
||||
await self.stop_ttfb_metrics()
|
||||
|
||||
frame = TTSAudioRawFrame(
|
||||
audio=resp.audio,
|
||||
sample_rate=self._sample_rate,
|
||||
num_channels=1,
|
||||
)
|
||||
yield frame
|
||||
|
||||
await self.start_tts_usage_metrics(text)
|
||||
yield TTSStoppedFrame()
|
||||
|
||||
|
||||
class ParakeetSTTService(STTService):
|
||||
class InputParams(BaseModel):
|
||||
language: Optional[Language] = Language.EN_US
|
||||
|
||||
def __init__(
|
||||
self,
|
||||
*,
|
||||
api_key: str,
|
||||
server: str = "grpc.nvcf.nvidia.com:443",
|
||||
function_id: str = "1598d209-5e27-4d3c-8079-4751568b1081",
|
||||
params: InputParams = InputParams(),
|
||||
**kwargs,
|
||||
):
|
||||
super().__init__(**kwargs)
|
||||
self._api_key = api_key
|
||||
self._profanity_filter = False
|
||||
self._automatic_punctuation = False
|
||||
self._no_verbatim_transcripts = False
|
||||
self._language_code = params.language
|
||||
self._boosted_lm_words = None
|
||||
self._boosted_lm_score = 4.0
|
||||
self._start_history = -1
|
||||
self._start_threshold = -1.0
|
||||
self._stop_history = -1
|
||||
self._stop_threshold = -1.0
|
||||
self._stop_history_eou = -1
|
||||
self._stop_threshold_eou = -1.0
|
||||
self._custom_configuration = ""
|
||||
self._sample_rate: int = 16000
|
||||
|
||||
self.set_model_name("parakeet-ctc-1.1b-asr")
|
||||
|
||||
metadata = [
|
||||
["function-id", function_id],
|
||||
["authorization", f"Bearer {api_key}"],
|
||||
]
|
||||
auth = riva.client.Auth(None, True, server, metadata)
|
||||
|
||||
self._asr_service = riva.client.ASRService(auth)
|
||||
|
||||
config = riva.client.StreamingRecognitionConfig(
|
||||
config=riva.client.RecognitionConfig(
|
||||
encoding=riva.client.AudioEncoding.LINEAR_PCM,
|
||||
language_code=self._language_code,
|
||||
model="",
|
||||
max_alternatives=1,
|
||||
profanity_filter=self._profanity_filter,
|
||||
enable_automatic_punctuation=self._automatic_punctuation,
|
||||
verbatim_transcripts=not self._no_verbatim_transcripts,
|
||||
sample_rate_hertz=self._sample_rate,
|
||||
audio_channel_count=1,
|
||||
),
|
||||
interim_results=True,
|
||||
)
|
||||
riva.client.add_word_boosting_to_config(
|
||||
config, self._boosted_lm_words, self._boosted_lm_score
|
||||
)
|
||||
riva.client.add_endpoint_parameters_to_config(
|
||||
config,
|
||||
self._start_history,
|
||||
self._start_threshold,
|
||||
self._stop_history,
|
||||
self._stop_history_eou,
|
||||
self._stop_threshold,
|
||||
self._stop_threshold_eou,
|
||||
)
|
||||
riva.client.add_custom_configuration_to_config(config, self._custom_configuration)
|
||||
self._config = config
|
||||
|
||||
self._queue = asyncio.Queue()
|
||||
|
||||
def can_generate_metrics(self) -> bool:
|
||||
return False
|
||||
|
||||
async def start(self, frame: StartFrame):
|
||||
await super().start(frame)
|
||||
self._thread_task = self.get_event_loop().create_task(self._thread_task_handler())
|
||||
self._response_task = self.get_event_loop().create_task(self._response_task_handler())
|
||||
self._response_queue = asyncio.Queue()
|
||||
|
||||
async def stop(self, frame: EndFrame):
|
||||
await super().stop(frame)
|
||||
await self._stop_tasks()
|
||||
|
||||
async def cancel(self, frame: CancelFrame):
|
||||
await super().cancel(frame)
|
||||
await self._stop_tasks()
|
||||
|
||||
async def _stop_tasks(self):
|
||||
self._thread_task.cancel()
|
||||
await self._thread_task
|
||||
self._response_task.cancel()
|
||||
await self._response_task
|
||||
|
||||
def _response_handler(self):
|
||||
responses = self._asr_service.streaming_response_generator(
|
||||
audio_chunks=self,
|
||||
streaming_config=self._config,
|
||||
)
|
||||
for response in responses:
|
||||
if not response.results:
|
||||
continue
|
||||
asyncio.run_coroutine_threadsafe(
|
||||
self._response_queue.put(response), self.get_event_loop()
|
||||
)
|
||||
|
||||
async def _thread_task_handler(self):
|
||||
try:
|
||||
self._thread_running = True
|
||||
await asyncio.to_thread(self._response_handler)
|
||||
except asyncio.CancelledError:
|
||||
self._thread_running = False
|
||||
pass
|
||||
|
||||
async def _handle_response(self, response):
|
||||
for result in response.results:
|
||||
if result and not result.alternatives:
|
||||
continue
|
||||
|
||||
transcript = result.alternatives[0].transcript
|
||||
if transcript and len(transcript) > 0:
|
||||
await self.stop_ttfb_metrics()
|
||||
if result.is_final:
|
||||
await self.stop_processing_metrics()
|
||||
await self.push_frame(
|
||||
TranscriptionFrame(transcript, "", time_now_iso8601(), None)
|
||||
)
|
||||
else:
|
||||
await self.push_frame(
|
||||
InterimTranscriptionFrame(transcript, "", time_now_iso8601(), None)
|
||||
)
|
||||
|
||||
async def _response_task_handler(self):
|
||||
while True:
|
||||
try:
|
||||
response = await self._response_queue.get()
|
||||
await self._handle_response(response)
|
||||
except asyncio.CancelledError:
|
||||
break
|
||||
|
||||
async def run_stt(self, audio: bytes) -> AsyncGenerator[Frame, None]:
|
||||
await self._queue.put(audio)
|
||||
yield None
|
||||
|
||||
def __next__(self) -> bytes:
|
||||
if not self._thread_running:
|
||||
raise StopIteration
|
||||
future = asyncio.run_coroutine_threadsafe(self._queue.get(), self.get_event_loop())
|
||||
return future.result()
|
||||
|
||||
def __iter__(self):
|
||||
return self
|
||||
@@ -7,8 +7,8 @@ deepgram-sdk~=3.5.0
|
||||
fal-client~=0.4.1
|
||||
fastapi~=0.115.0
|
||||
faster-whisper~=1.0.3
|
||||
google-cloud-texttospeech~=2.17.2
|
||||
google-generativeai~=0.7.2
|
||||
google-cloud-texttospeech~=2.21.1
|
||||
google-generativeai~=0.8.3
|
||||
langchain~=0.2.14
|
||||
livekit~=0.13.1
|
||||
lmnt~=1.1.4
|
||||
|
||||
Reference in New Issue
Block a user