Add torch compile to log mel spectrogram

whisperx/asr.py

@@ -181,6 +181,9 @@ class FasterWhisperPipeline(Pipeline):
 
     def preprocess(self, audio):
         audio = audio['inputs']
+        if isinstance(audio, np.ndarray):
+            audio = torch.from_numpy(audio)
+
         features = log_mel_spectrogram(audio, padding=N_SAMPLES - audio.shape[0])
         return {'inputs': features}
 
@@ -253,7 +256,7 @@ class FasterWhisperPipeline(Pipeline):
     def detect_language(self, audio: np.ndarray):
         if audio.shape[0] < N_SAMPLES:
             print("Warning: audio is shorter than 30s, language detection may be inaccurate.")
-        segment = log_mel_spectrogram(audio[: N_SAMPLES],
+        segment = log_mel_spectrogram(torch.from_numpy(audio[:N_SAMPLES]),
                                       padding=0 if audio.shape[0] >= N_SAMPLES else N_SAMPLES - audio.shape[0])
         encoder_output = self.model.encode(segment)
         results = self.model.model.detect_language(encoder_output)

whisperx/audio.py

@@ -22,6 +22,12 @@ N_SAMPLES_PER_TOKEN = HOP_LENGTH * 2  # the initial convolutions has stride 2
 FRAMES_PER_SECOND = exact_div(SAMPLE_RATE, HOP_LENGTH)  # 10ms per audio frame
 TOKENS_PER_SECOND = exact_div(SAMPLE_RATE, N_SAMPLES_PER_TOKEN)  # 20ms per audio token
 
+with np.load(
+    os.path.join(os.path.dirname(__file__), "assets", "mel_filters.npz")
+) as f:
+    MEL_FILTERS = torch.from_numpy(f[f"mel_{80}"])
+
+
 
 def load_audio(file: str, sr: int = SAMPLE_RATE):
     """
@@ -79,27 +85,9 @@ def pad_or_trim(array, length: int = N_SAMPLES, *, axis: int = -1):
     return array
 
 
-@lru_cache(maxsize=None)
+@torch.compile(fullgraph=True)
-def mel_filters(device, n_mels: int = N_MELS) -> torch.Tensor:
-    """
-    load the mel filterbank matrix for projecting STFT into a Mel spectrogram.
-    Allows decoupling librosa dependency; saved using:
-
-        np.savez_compressed(
-            "mel_filters.npz",
-            mel_80=librosa.filters.mel(sr=16000, n_fft=400, n_mels=80),
-        )
-    """
-    assert n_mels == 80, f"Unsupported n_mels: {n_mels}"
-    with np.load(
-        os.path.join(os.path.dirname(__file__), "assets", "mel_filters.npz")
-    ) as f:
-        return torch.from_numpy(f[f"mel_{n_mels}"]).to(device)
-
-
 def log_mel_spectrogram(
-    audio: Union[str, np.ndarray, torch.Tensor],
+    audio: torch.Tensor,
-    n_mels: int = N_MELS,
     padding: int = 0,
     device: Optional[Union[str, torch.device]] = None,
 ):
@@ -108,7 +96,7 @@ def log_mel_spectrogram(
 
     Parameters
     ----------
-    audio: Union[str, np.ndarray, torch.Tensor], shape = (*)
+    audio: torch.Tensor, shape = (*)
         The path to audio or either a NumPy array or Tensor containing the audio waveform in 16 kHz
 
     n_mels: int
@@ -125,21 +113,19 @@ def log_mel_spectrogram(
     torch.Tensor, shape = (80, n_frames)
         A Tensor that contains the Mel spectrogram
     """
-    if not torch.is_tensor(audio):
+    global MEL_FILTERS
-        if isinstance(audio, str):
-            audio = load_audio(audio)
-        audio = torch.from_numpy(audio)
 
     if device is not None:
         audio = audio.to(device)
     if padding > 0:
         audio = F.pad(audio, (0, padding))
     window = torch.hann_window(N_FFT).to(audio.device)
-    stft = torch.stft(audio, N_FFT, HOP_LENGTH, window=window, return_complex=True)
+    stft = torch.stft(audio, N_FFT, HOP_LENGTH, window=window, return_complex=False)
-    magnitudes = stft[..., :-1].abs() ** 2
+    # Square the real and imaginary components and sum them together, similar to torch.abs() on complex tensors
+    magnitudes = (stft[:, :-1, :] ** 2).sum(dim=-1)
 
-    filters = mel_filters(audio.device, n_mels)
+    MEL_FILTERS = MEL_FILTERS.to(audio.device)
-    mel_spec = filters @ magnitudes
+    mel_spec = MEL_FILTERS @ magnitudes
 
     log_spec = torch.clamp(mel_spec, min=1e-10).log10()
     log_spec = torch.maximum(log_spec, log_spec.max() - 8.0)