tested

backend/routers/transcribe.py

@@ -15,7 +15,7 @@
 import opuslib
 
 from utils.stt.deepgram_util import transcribe_file_deepgram, process_audio_dg, send_initial_file, \
-    get_speaker_audio_file, convert_audio_bytes_to_resampled_bytes
+    get_speaker_audio_file
 from utils.stt.vad import vad_is_empty, VADIterator, model, is_speech_present
 
 router = APIRouter()
@@ -75,7 +75,7 @@ async def _websocket_util(
     transcript_socket2 = None
     duration = 0
     is_speech_active = False
-    speech_timeout = 0.7
+    speech_timeout = 0.7 # Good for now but better dynamically adjust it by user behaviour
     last_speech_time = 0
     try:
         # single_file_path, duration = get_speaker_audio_file(uid) if language == 'en' else (None, 0)
@@ -93,7 +93,7 @@ async def _websocket_util(
         await websocket.close()
         return
 
-    threshold = 0.7
+    threshold = 0.6
     vad_iterator = VADIterator(model, sampling_rate=sample_rate, threshold=threshold) 
     window_size_samples = 256 if sample_rate == 8000 else 512
     if codec == 'opus':