Add --live_input_device_sample_rate parameter

src/whisper_ctranslate2/commandline.py

@@ -446,4 +446,11 @@ def read_command_line():
             help="Set live stream input device ID (see python -m sounddevice for a list)",
         )
 
+        live_args.add_argument(
+            "--live_input_device_sample_rate",
+            type=int,
+            default=16000,
+            help="Set live sample rate of input device",
+        )
+
         return parser.parse_args().__dict__

src/whisper_ctranslate2/live.py

@@ -6,7 +6,6 @@
 
 from .transcribe import Transcribe, TranscriptionOptions
 
-SampleRate = 16000  # Stream device recording frequency per second
 BlockSize = 30  # Block size in milliseconds
 Vocals = [50, 1000]  # Frequency range to detect sounds that could be speech
 EndBlocks = 33 * 2  # Number of blocks to wait before sending (30 ms is block)
@@ -38,6 +37,7 @@ def __init__(
         verbose: bool,
         threshold: float,
         input_device: int,
+        input_device_sample_rate: int,
         options: TranscriptionOptions,
     ):
         self.model_path = model_path
@@ -52,6 +52,7 @@ def __init__(
         self.verbose = verbose
         self.threshold = threshold
         self.input_device = input_device
+        self.input_device_sample_rate = input_device_sample_rate
         self.options = options
 
         self.running = True
@@ -71,7 +72,11 @@ def force_not_available_exception():
         raise (sounddevice_exception)
 
     def _is_there_voice(self, indata, frames):
-        freq = np.argmax(np.abs(np.fft.rfft(indata[:, 0]))) * SampleRate / frames
+        freq = (
+            np.argmax(np.abs(np.fft.rfft(indata[:, 0])))
+            * self.input_device_sample_rate
+            / frames
+        )
         volume = np.sqrt(np.mean(indata**2))
 
         return volume > self.threshold and Vocals[0] <= freq <= Vocals[1]
@@ -158,8 +163,8 @@ def listen(self):
         with sd.InputStream(
             channels=1,
             callback=self.callback,
-            blocksize=int(SampleRate * BlockSize / 1000),
-            samplerate=SampleRate,
+            blocksize=int(self.input_device_sample_rate * BlockSize / 1000),
+            samplerate=self.input_device_sample_rate,
             device=self.input_device,
         ):
             while self.running:

src/whisper_ctranslate2/whisper_ctranslate2.py

@@ -114,6 +114,7 @@ def main():
     local_files_only: bool = args.pop("local_files_only")
     live_volume_threshold: float = args.pop("live_volume_threshold")
     live_input_device: int = args.pop("live_input_device")
+    live_input_device_sample_rate: int = args.pop("live_input_device_sample_rate")
     hf_token = args.pop("hf_token")
     speaker_name = args.pop("speaker_name")
     batched = args.pop("batched")
@@ -207,6 +208,7 @@ def main():
             verbose,
             live_volume_threshold,
             live_input_device,
+            live_input_device_sample_rate,
             options,
         ).inference()
 

tests/testlive.py

@@ -19,6 +19,7 @@ def test_constructor(self):
             threshold=0.2,
             input_device=0,
             options=None,
+            input_device_sample_rate=16000,
         )
         self.assertNotEqual(None, live)