feat: add multi-channel support and language filtering to focus on specific speakers

README.md

@@ -65,6 +65,10 @@ Run the script using `python3 transcribe.py` with optional flags.
 - `-c`, `--context`: Enable prompt caching/rolling context to help the model maintain sentence continuity across chunks.
 - `--lang [CODE]`: Hardcode the source language (e.g., `en`, `es`) to bypass automatic language detection for faster processing.
 - `--silence [MS]`: Set the minimum silence duration in milliseconds to end a chunk. Defaults to 1000ms. Increase to force longer sentences before translation.
+- `--max-buffer [SEC]`: Maximum buffer duration in seconds before forcing a flush (default: 20).
+- `--channels [N]`: Number of input channels (default: 1).
+- `--pick-channel [0|1]`: If stereo, select channel 0 (Left) or 1 (Right) to focus transcription.
+- `--filter-lang`: If used with `--lang`, discards segments that do not match the target language.
 
 ---
 

transcribe.py

@@ -43,7 +43,6 @@ TARGET_LANGS = {
     "ar": "Helsinki-NLP/opus-mt-en-ar" # Added Arabic as discussed before
 }
 
-CHANNELS = 1
 SAMPLERATE = 16000
 BLOCK_SIZE = 512
 VAD_THRESHOLD = 0.5 
@@ -127,6 +126,9 @@ def main():
     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)")
     parser.add_argument("--max-buffer", type=int, default=20, help="Maximum buffer duration in seconds before forcing a flush (default: 20)")
+    parser.add_argument("--channels", type=int, default=1, help="Number of input channels (default: 1)")
+    parser.add_argument("--pick-channel", type=int, choices=[0, 1], help="Pick a specific channel (0 or 1) from stereo input")
+    parser.add_argument("--filter-lang", action="store_true", help="Discard segments where detected language does not match --lang")
     
     args = parser.parse_args()
 
@@ -210,10 +212,17 @@ def main():
     last_draft_text = ""
 
     try:
-        with sd.InputStream(samplerate=SAMPLERATE, channels=CHANNELS, callback=callback, blocksize=BLOCK_SIZE, device=device_index):
+        with sd.InputStream(samplerate=SAMPLERATE, channels=args.channels, callback=callback, blocksize=BLOCK_SIZE, device=device_index):
             while True:
                 while not audio_queue.empty():
                     data = audio_queue.get()
+                    if args.channels > 1:
+                        if args.pick_channel is not None:
+                            # Select specific channel
+                            data = data[:, args.pick_channel]
+                        else:
+                            # Mix to mono
+                            data = np.mean(data, axis=1)
                     audio_buffer.append(data.flatten())
 
                 if len(audio_buffer) > 0:
@@ -269,6 +278,11 @@ def main():
                             original_text = transcription_result['text'].strip()
                             detected_lang = transcription_result.get('language', args.lang if args.lang else 'en')
                             
+                            # Filter language if requested
+                            if args.filter_lang and args.lang and detected_lang != args.lang:
+                                print(f"\n[SYSTEM]: Discarding segment (Detected: {detected_lang}, Expected: {args.lang})")
+                                original_text = ""
+
                             if is_hallucination(original_text):
                                 print(f"\n[SYSTEM]: Hallucination detected, ignoring and resetting context.")
                                 original_text = ""
@@ -344,7 +358,11 @@ def main():
                             
                             draft_result = mlx_whisper.transcribe(current_audio, **draft_kwargs)
                             draft_text = draft_result['text'].strip()
-                            
+                            draft_lang = draft_result.get('language', args.lang if args.lang else 'en')
+
+                            if args.filter_lang and args.lang and draft_lang != args.lang:
+                                draft_text = ""
+
                             if draft_text:
                                 # Update timestamp ONLY if the text actually changed
                                 if draft_text != last_draft_text: