admin-valentin/WhisperLiveKit
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Refactor PCM conversion to a dedicated function; immediate chunk addition to the diarization queue

Browse source
This commit is contained in:
Quentin Fuxa 2025-03-17 11:46:45 +01:00
parent eead544977
commit 4f87ac3ea4
2 changed files with 18 additions and 9 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

4
README.md
View file

@ -30,8 +30,8 @@ This project is based on [Whisper Streaming](https://github.com/ufal/whisper_str
1. **Clone the Repository**: 1. **Clone the Repository**:
```bash ```bash
git clone https://github.com/QuentinFuxa/whisper_streaming_web git clone https://github.com/QuentinFuxa/WhisperLiveKit
cd whisper_streaming_web cd WhisperLiveKit
``` ```

23
whisper_fastapi_online_server.py
View file

@ -190,6 +190,16 @@ app.add_middleware(
with open("web/live_transcription.html", "r", encoding="utf-8") as f: with open("web/live_transcription.html", "r", encoding="utf-8") as f:
html = f.read() html = f.read()
def convert_pcm_to_float(pcm_buffer):
"""
Converts a PCM buffer in s16le format to a normalized NumPy array.
Arg: pcm_buffer. PCM buffer containing raw audio data in s16le format
Returns: np.ndarray. NumPy array of float32 type normalized between -1.0 and 1.0
"""
pcm_array = (np.frombuffer(pcm_buffer, dtype=np.int16).astype(np.float32)
/ 32768.0)
return pcm_array
async def start_ffmpeg_decoder(): async def start_ffmpeg_decoder():
""" """
Start an FFmpeg process in async streaming mode that reads WebM from stdin Start an FFmpeg process in async streaming mode that reads WebM from stdin
@ -444,24 +454,23 @@ async def websocket_endpoint(websocket: WebSocket):
logger.info("FFmpeg stdout closed.") logger.info("FFmpeg stdout closed.")
break break
pcm_buffer.extend(chunk) pcm_buffer.extend(chunk)
if args.diarization and diarization_queue:
await diarization_queue.put(convert_pcm_to_float(pcm_buffer).copy())
if len(pcm_buffer) >= BYTES_PER_SEC: if len(pcm_buffer) >= BYTES_PER_SEC:
if len(pcm_buffer) > MAX_BYTES_PER_SEC: if len(pcm_buffer) > MAX_BYTES_PER_SEC:
logger.warning( logger.warning(
f"""Audio buffer is too large: {len(pcm_buffer) / BYTES_PER_SEC:.2f} seconds. f"""Audio buffer is too large: {len(pcm_buffer) / BYTES_PER_SEC:.2f} seconds.
The model probably struggles to keep up. Consider using a smaller model. The model probably struggles to keep up. Consider using a smaller model.
""") """)
# Convert int16 -> float32
pcm_array = ( pcm_array = convert_pcm_to_float(pcm_buffer[:MAX_BYTES_PER_SEC])
np.frombuffer(pcm_buffer[:MAX_BYTES_PER_SEC], dtype=np.int16).astype(np.float32)
/ 32768.0
)
pcm_buffer = pcm_buffer[MAX_BYTES_PER_SEC:] pcm_buffer = pcm_buffer[MAX_BYTES_PER_SEC:]
if args.transcription and transcription_queue: if args.transcription and transcription_queue:
await transcription_queue.put(pcm_array.copy()) await transcription_queue.put(pcm_array.copy())
if args.diarization and diarization_queue:
await diarization_queue.put(pcm_array.copy())
if not args.transcription and not args.diarization: if not args.transcription and not args.diarization:
await asyncio.sleep(0.1) await asyncio.sleep(0.1)