From 6f84597da33557692dbf7d20a62e91142e6717aa Mon Sep 17 00:00:00 2001
From: Subliminal Guy <subliminal_kid@posteo.de>
Date: Mon, 21 Apr 2025 16:35:07 +0200
Subject: [PATCH] Make Transcription async so that Status Route stays accesible
---
app/webservice.py | 27 +++++++++++++++++----------
1 file changed, 17 insertions(+), 10 deletions(-)
diff --git a/app/webservice.py b/app/webservice.py
index 38431fa..d0c39b9 100644
--- a/app/webservice.py
+++ b/app/webservice.py
@@ -11,6 +11,7 @@ from fastapi.openapi.docs import get_swagger_ui_html
from fastapi.responses import RedirectResponse, StreamingResponse
from fastapi.staticfiles import StaticFiles
from whisper import tokenizer
+import asyncio
from app.config import CONFIG
from app.factory.asr_model_factory import ASRModelFactory
@@ -88,16 +89,22 @@ async def asr(
),
output: Union[str, None] = Query(default="txt", enum=["txt", "vtt", "srt", "tsv", "json"]),
):
- result = asr_model.transcribe(
- load_audio(audio_file.file, encode),
- task,
- language,
- initial_prompt,
- vad_filter,
- word_timestamps,
- {"diarize": diarize, "min_speakers": min_speakers, "max_speakers": max_speakers},
- output,
- )
+ # Run transcription in a background thread to keep the event loop responsive
+ def _run_transcription():
+ audio = load_audio(audio_file.file, encode)
+ return asr_model.transcribe(
+ audio,
+ task,
+ language,
+ initial_prompt,
+ vad_filter,
+ word_timestamps,
+ {"diarize": diarize, "min_speakers": min_speakers, "max_speakers": max_speakers},
+ output,
+ )
+ # offload blocking transcription to a thread
+ result = await asyncio.to_thread(_run_transcription)
+ # stream the transcription result back to the client
return StreamingResponse(
result,
media_type="text/plain",
--
GitLab