Logan
71 lines
2.3 KiB
Python
71 lines
2.3 KiB
Python
"""
|
|
Speech-to-text transcription for recorded calls.
|
|
|
|
Uses Google Cloud Speech-to-Text v1 (authenticated via the same ADC / service
|
|
account used by firebase-admin and google-cloud-storage).
|
|
|
|
Triggered as a background task from the upload endpoint after a call audio
|
|
file has been successfully stored in GCS.
|
|
"""
|
|
import asyncio
|
|
from typing import Optional
|
|
from app.internal.logger import logger
|
|
from app.internal import firestore as fstore
|
|
|
|
|
|
async def transcribe_call(call_id: str, gcs_uri: str) -> Optional[str]:
|
|
"""
|
|
Transcribe audio at the given GCS URI and store the result in Firestore.
|
|
|
|
Args:
|
|
call_id: Firestore document ID in the 'calls' collection.
|
|
gcs_uri: GCS URI of the audio file, e.g. gs://bucket/calls/xyz.mp3
|
|
|
|
Returns:
|
|
The transcript string, or None if transcription failed / was skipped.
|
|
"""
|
|
if not gcs_uri or not gcs_uri.startswith("gs://"):
|
|
return None
|
|
|
|
try:
|
|
transcript = await asyncio.to_thread(_sync_transcribe, gcs_uri)
|
|
except Exception as e:
|
|
logger.warning(f"Transcription failed for call {call_id}: {e}")
|
|
return None
|
|
|
|
if transcript:
|
|
try:
|
|
await fstore.doc_update("calls", call_id, {"transcript": transcript})
|
|
logger.info(f"Transcript saved for call {call_id} ({len(transcript)} chars)")
|
|
except Exception as e:
|
|
logger.warning(f"Could not save transcript for {call_id}: {e}")
|
|
|
|
return transcript
|
|
|
|
|
|
def _sync_transcribe(gcs_uri: str) -> Optional[str]:
|
|
"""Synchronous STT call — run in a thread via asyncio.to_thread."""
|
|
from google.cloud import speech
|
|
|
|
client = speech.SpeechClient()
|
|
|
|
audio = speech.RecognitionAudio(uri=gcs_uri)
|
|
config = speech.RecognitionConfig(
|
|
encoding=speech.RecognitionConfig.AudioEncoding.MP3,
|
|
sample_rate_hertz=22050,
|
|
language_code="en-US",
|
|
enable_automatic_punctuation=True,
|
|
model="latest_long",
|
|
)
|
|
|
|
# Use long_running_recognize for reliability; it handles both short and long audio
|
|
operation = client.long_running_recognize(config=config, audio=audio)
|
|
response = operation.result(timeout=120)
|
|
|
|
parts = [
|
|
result.alternatives[0].transcript
|
|
for result in response.results
|
|
if result.alternatives
|
|
]
|
|
return " ".join(parts).strip() or None
|