Fix speech profile but only for pcm8 and english

app/lib/pages/capture/page.dart

@@ -81,6 +81,7 @@ class CapturePageState extends State<CapturePage>
 
   Future<void> initiateWebsocket([BleAudioCodec? audioCodec, int? sampleRate]) async {
     BleAudioCodec codec = audioCodec ?? (btDevice?.id == null ? BleAudioCodec.pcm8 : await getAudioCodec(btDevice!.id));
+    int sampleRate = (codec == BleAudioCodec.opus ? 16000 : 8000);
     await initWebSocket(
       codec: codec,
       sampleRate: sampleRate,

backend/routers/transcribe.py

@@ -72,13 +72,10 @@ async def _websocket_util(
     transcript_socket2 = None
     duration = 0
     try:
-        # if language == 'en':
-        #     # TODO: if pcm16? meaning phone recording, should ignore?
-        #     single_file_path, duration = get_speaker_audio_file(uid, target_sample_rate=sample_rate)
-        # else:
-        #     single_file_path, duration = None, 0
-        single_file_path, duration = None, 0
-        # TODO: what if opus? and the samples were created with pcm?
+        if language == 'en' and codec == 'pcm8':  # no pcm16 which is phone recording, no opus
+            single_file_path, duration = get_speaker_audio_file(uid, target_sample_rate=sample_rate)
+        else:
+            single_file_path, duration = None, 0
         transcript_socket = await process_audio_dg(websocket, language, sample_rate, codec, channels,
                                                    preseconds=duration)
         if duration:
@@ -118,7 +115,7 @@ async def receive_audio(socket1, socket2):
                 #         continue
                 #
                 #     audio_buffer = audio_buffer[window_size_samples * 2:]
-
+                # print(data)
                 elapsed_seconds = time.time() - timer_start
                 if elapsed_seconds > duration or not socket2:
                     socket1.send(data)

backend/utils/storage.py

@@ -59,7 +59,7 @@ def retrieve_all_samples(uid: str):
 
     for i, blob in enumerate(blobs):
         path = f'{base_path}{blob.name.split("/")[-1]}'
-        if os.path.exists(path):
+        if os.path.exists(path):  # when opus uploaded? should refresh the download
             continue
         try:
             blob.download_to_filename(path)

backend/utils/stt/deepgram_util.py

@@ -93,14 +93,17 @@ def get_single_file(dir_path: str, target_sample_rate: int = 8000):
     for sample in os.listdir(dir_path):
         if '.wav' not in sample:
             continue
+        if 'joined_output.wav' in sample:
+            continue
         path = f'{dir_path}{sample}'
         aseg = AudioSegment.from_file(path)
-        if aseg.frame_rate != target_sample_rate:
-            print('converting', aseg.frame_rate, 'to', target_sample_rate)
-            aseg = aseg.set_frame_rate(target_sample_rate)
-        files_to_join.append(aseg)
+        print(path, aseg.frame_rate)
+        if aseg.frame_rate == target_sample_rate:
+            files_to_join.append(aseg)
 
-    output = files_to_join[0]  # KNOWN ISSUE
+    if not files_to_join:
+        return None
+    output = files_to_join[0]
     for audio in files_to_join[1:]:
         output += audio
 
@@ -117,17 +120,17 @@ def get_speaker_audio_file(uid: str, target_sample_rate: int = 8000) -> Tuple[Op
     print('get_speaker_audio_file', uid, path, 'Files:', files_at_path)
     if files_at_path < 5:  # means user did less than 5 samples unfortunately, so not completed
         return None, 0
-
     single_file_path = f'{path}joined_output.wav'
     if os.path.exists(single_file_path):
         aseg = AudioSegment.from_wav(single_file_path)
-        if aseg.frame_rate == target_sample_rate:
-            aseg = aseg.set_frame_rate(target_sample_rate)
-            aseg.export(single_file_path, format='wav')
-        print('get_speaker_audio_file Cached Duration:', aseg.duration_seconds)
-        return single_file_path, aseg.duration_seconds
+        if aseg.frame_rate == target_sample_rate:  # sample sample rate
+            print('get_speaker_audio_file Cached Duration:', aseg.duration_seconds)
+            return single_file_path, aseg.duration_seconds
 
     single_file_path = get_single_file(path, target_sample_rate)
+    if not single_file_path:
+        return None, 0  # no files for this codec
+
     aseg = AudioSegment.from_wav(single_file_path)
     print('get_speaker_audio_file Initial Duration:', aseg.duration_seconds, 'Sample rate:', aseg.frame_rate / 1000)
     output = AudioSegment.empty()
@@ -137,14 +140,14 @@ def get_speaker_audio_file(uid: str, target_sample_rate: int = 8000) -> Tuple[Op
         end = segment['end'] * 1000
         output += aseg[start:end]
 
-    if output.duration_seconds < 20:
+    seconds = 30
+    if output.duration_seconds < seconds:
         print('get_speaker_audio_file Output Duration:', output.duration_seconds)
-        return single_file_path, output.duration_seconds
+        return single_file_path, output.duration_seconds  # < 30
 
-    seconds = 20
-    output = output[:20 * 1000]
+    output = output[:seconds * 1000]
     output.export(single_file_path, format="wav")
-    return single_file_path, seconds
+    return single_file_path, output.duration_seconds  # 30
 
 
 deepgram = DeepgramClient(os.getenv('DEEPGRAM_API_KEY'), DeepgramClientOptions(options={"keepalive": "true"}))
@@ -156,7 +159,7 @@ async def process_audio_dg(
     loop = asyncio.get_event_loop()
 
     def on_message(self, result, **kwargs):
-        print("Received message from Deepgram")  # Log when message is received
+        print(f"Received message from Deepgram")  # Log when message is received
         sentence = result.channel.alternatives[0].transcript
         if len(sentence) == 0:
             return