Improve speaker handling; update sleep duration and manage speaker transitions more effectively

This commit is contained in:
Quentin Fuxa 2025-02-19 14:41:37 +01:00
parent 58eba2a1f6
commit 2608abf0f3

View file

@ -214,10 +214,10 @@ async def websocket_endpoint(websocket: WebSocket):
else: else:
chunk_history.append({ chunk_history.append({
"beg": time() - beg_loop, "beg": time() - beg_loop,
"end": time() - beg_loop + 0.1, "end": time() - beg_loop + 1,
"text": '', "text": '',
}) })
sleep(0.1) sleep(1)
buffer = '' buffer = ''
if args.diarization: if args.diarization:
@ -225,28 +225,29 @@ async def websocket_endpoint(websocket: WebSocket):
diarization.assign_speakers_to_chunks(chunk_history) diarization.assign_speakers_to_chunks(chunk_history)
current_speaker = -1 current_speaker = 0
lines = [{ lines = []
"beg": 0, last_end_diarized = 0
"end": 0, for ind, ch in enumerate(chunk_history):
"speaker": current_speaker, speaker = ch.get("speaker", -3)
"text": "" if speaker == -1 and ind < len(chunk_history) - 1:
}] continue
for ch in chunk_history: elif speaker != current_speaker:
if args.diarization and ch["speaker"] and ch["speaker"] != current_speaker:
new_speaker = ch["speaker"]
lines.append( lines.append(
{ {
"speaker": new_speaker, "speaker": speaker,
"text": ch['text'], "text": ch['text'],
"beg": format_time(ch['beg']), "beg": format_time(ch['beg']),
"end": format_time(ch['end']), "end": format_time(ch['end']),
"diff": round(ch['end'] - last_end_diarized, 2)
} }
) )
current_speaker = new_speaker current_speaker = speaker
else: elif speaker != -1:
lines[-1]["text"] += ch['text'] lines[-1]["text"] += ch['text']
lines[-1]["end"] = format_time(ch['end']) lines[-1]["end"] = format_time(ch['end'])
if speaker != -1:
last_end_diarized = max(ch['end'], last_end_diarized)
response = {"lines": lines, "buffer": buffer} response = {"lines": lines, "buffer": buffer}
await websocket.send_json(response) await websocket.send_json(response)