Commit @dc7bb9d51ff1e10178144b51762f168491886dec - yjyoon/whisper_server

Fedir Zadniprovskyi 2024-12-17

feat: add chat, speech, and transcription dependencies

@dc7bb9d51ff1e10178144b51762f168491886dec

ed63191

dc7bb9d

src/faster_whisper_server/config.py

--- src/faster_whisper_server/config.py

+++ src/faster_whisper_server/config.py

...	...	@@ -236,3 +236,14 @@
236	236	Controls how many latest seconds of audio are being passed through VAD.
237	237	Should be greater than `max_inactivity_seconds`
238	238	"""
	239	+
	240	+ chat_completion_base_url: str = "https://api.openai.com/v1"
	241	+ chat_completion_api_key: str \| None = None
	242	+
	243	+ speech_base_url: str \| None = None
	244	+ speech_api_key: str \| None = None
	245	+ speech_model: str = "piper"
	246	+ speech_extra_body: dict = {"sample_rate": 24000}
	247	+
	248	+ transcription_base_url: str \| None = None
	249	+ transcription_api_key: str \| None = None

ed63191

dc7bb9d

src/faster_whisper_server/dependencies.py

--- src/faster_whisper_server/dependencies.py

+++ src/faster_whisper_server/dependencies.py

...	...	@@ -3,6 +3,10 @@
3	3
4	4	from fastapi import Depends, HTTPException, status
5	5	from fastapi.security import HTTPAuthorizationCredentials, HTTPBearer
	6	+from httpx import ASGITransport, AsyncClient
	7	+from openai import AsyncOpenAI
	8	+from openai.resources.audio import AsyncSpeech, AsyncTranscriptions
	9	+from openai.resources.chat.completions import AsyncCompletions
6	10
7	11	from faster_whisper_server.config import Config
8	12	from faster_whisper_server.model_manager import PiperModelManager, WhisperModelManager
...	...	@@ -45,3 +49,56 @@
45	49
46	50
47	51	ApiKeyDependency = Depends(verify_api_key)
	52	+
	53	+
	54	+@lru_cache
	55	+def get_completion_client() -> AsyncCompletions:
	56	+ config = get_config() # HACK
	57	+ oai_client = AsyncOpenAI(base_url=config.chat_completion_base_url, api_key=config.chat_completion_api_key)
	58	+ return oai_client.chat.completions
	59	+
	60	+
	61	+CompletionClientDependency = Annotated[AsyncCompletions, Depends(get_completion_client)]
	62	+
	63	+
	64	+@lru_cache
	65	+def get_speech_client() -> AsyncSpeech:
	66	+ config = get_config() # HACK
	67	+ if config.speech_base_url is None:
	68	+ # this might not work as expected if the `speech_router` won't have shared state with the main FastAPI `app`. TODO: verify # noqa: E501
	69	+ from faster_whisper_server.routers.speech import (
	70	+ router as speech_router,
	71	+ )
	72	+
	73	+ http_client = AsyncClient(
	74	+ transport=ASGITransport(speech_router), base_url="http://test/v1"
	75	+ ) # NOTE: "test" can be replaced with any other value
	76	+ oai_client = AsyncOpenAI(http_client=http_client, api_key=config.speech_api_key)
	77	+ else:
	78	+ oai_client = AsyncOpenAI(base_url=config.speech_base_url, api_key=config.speech_api_key)
	79	+ return oai_client.audio.speech
	80	+
	81	+
	82	+SpeechClientDependency = Annotated[AsyncSpeech, Depends(get_speech_client)]
	83	+
	84	+
	85	+@lru_cache
	86	+def get_transcription_client() -> AsyncTranscriptions:
	87	+ config = get_config()
	88	+ if config.transcription_base_url is None:
	89	+ # this might not work as expected if the `transcription_router` won't have shared state with the main FastAPI `app`. TODO: verify # noqa: E501
	90	+ from faster_whisper_server.routers.stt import (
	91	+ router as stt_router,
	92	+ )
	93	+
	94	+ http_client = AsyncClient(
	95	+ transport=ASGITransport(stt_router), base_url="http://test/v1"
	96	+ ) # NOTE: "test" can be replaced with any other value
	97	+
	98	+ oai_client = AsyncOpenAI(http_client=http_client, api_key=config.transcription_api_key)
	99	+ else:
	100	+ oai_client = AsyncOpenAI(base_url=config.transcription_base_url, api_key=config.transcription_api_key)
	101	+ return oai_client.audio.transcriptions
	102	+
	103	+
	104	+TranscriptionClientDependency = Annotated[AsyncTranscriptions, Depends(get_transcription_client)]

Add a comment

Open 0
Closed 0

List

Delete comment