feat: parameterize min silence duration and increase default to 1000ms

transcribe.py

@@ -46,7 +46,6 @@ SAMPLERATE = 16000
 BLOCK_SIZE = 512
 VAD_THRESHOLD = 0.5 
 BUFFER_LIMIT = SAMPLERATE * 30 
-MIN_SILENCE_DURATION_MS = 500 
 
 audio_queue = queue.Queue()
 ingest_queue = queue.Queue()
@@ -96,6 +95,7 @@ def main():
     parser.add_argument("-s", "--stream", action="store_true", help="Enable real-time streaming transcription (Draft mode)")
     parser.add_argument("-c", "--context", action="store_true", help="Enable prompt caching/rolling context for better continuity")
     parser.add_argument("--lang", type=str, help="Hardcode source language (e.g. 'en', 'es') to bypass detection")
+    parser.add_argument("--silence", type=int, default=1000, help="Minimum silence duration in ms to end a chunk (default: 1000)")
     
     args = parser.parse_args()
 
@@ -167,7 +167,7 @@ def main():
                         vad_model, 
                         sampling_rate=SAMPLERATE,
                         threshold=VAD_THRESHOLD,
-                        min_silence_duration_ms=MIN_SILENCE_DURATION_MS
+                        min_silence_duration_ms=args.silence
                     )
 
                     if len(speech_timestamps) > 0:
@@ -175,7 +175,7 @@ def main():
                         last_end = speech_timestamps[-1]['end']
                         buffer_len_samples = len(current_audio)
                         
-                        if (buffer_len_samples - last_end) > (SAMPLERATE * MIN_SILENCE_DURATION_MS / 1000) or buffer_len_samples > BUFFER_LIMIT:
+                        if (buffer_len_samples - last_end) > (SAMPLERATE * args.silence / 1000) or buffer_len_samples > BUFFER_LIMIT:
                             
                             # Clear draft line if it was used
                             if args.stream: