Update app.py

app.py

@@ -111,9 +111,7 @@ def demo_fn(speech_upl: str, noise_type: str, snr: int, mic_input: Optional[str]
     snr = int(snr)
     noise_fn = NOISES[noise_type]
     meta = AudioMetaData(-1, -1, -1, -1, "")
-
-    max_s = 3600  # allow up to 1 hour
-
+    max_s = 10  # limit to 10 seconds
     if speech_upl is not None:
         sample, meta = load_audio(speech_upl, sr)
         max_len = max_s * sr
@@ -123,51 +121,43 @@ def demo_fn(speech_upl: str, noise_type: str, snr: int, mic_input: Optional[str]
     else:
         sample, meta = load_audio("samples/p232_013_clean.wav", sr)
         sample = sample[..., : max_s * sr]
-
     if sample.dim() > 1 and sample.shape[0] > 1:
-        assert sample.shape[1] > sample.shape[0], f"Expecting channels first, but got {sample.shape}"
+        assert (
+            sample.shape[1] > sample.shape[0]
+        ), f"Expecting channels first, but got {sample.shape}"
         sample = sample.mean(dim=0, keepdim=True)
-
     logger.info(f"Loaded sample with shape {sample.shape}")
-
     if noise_fn is not None:
         noise, _ = load_audio(noise_fn, sr)  # type: ignore
         logger.info(f"Loaded noise with shape {noise.shape}")
         _, _, sample = mix_at_snr(sample, noise, snr)
-
     logger.info("Start denoising audio")
     enhanced = enhance(model, df, sample)
     logger.info("Denoising finished")
-
     lim = torch.linspace(0.0, 1.0, int(sr * 0.15)).unsqueeze(0)
     lim = torch.cat((lim, torch.ones(1, enhanced.shape[1] - lim.shape[1])), dim=1)
     enhanced = enhanced * lim
-
     if meta.sample_rate != sr:
         enhanced = resample(enhanced, sr, meta.sample_rate)
         sample = resample(sample, sr, meta.sample_rate)
         sr = meta.sample_rate
-
     noisy_wav = tempfile.NamedTemporaryFile(suffix="noisy.wav", delete=False).name
     save_audio(noisy_wav, sample, sr)
     enhanced_wav = tempfile.NamedTemporaryFile(suffix="enhanced.wav", delete=False).name
     save_audio(enhanced_wav, enhanced, sr)
-
     logger.info(f"saved audios: {noisy_wav}, {enhanced_wav}")
-
     ax_noisy.clear()
     ax_enh.clear()
     noisy_im = spec_im(sample, sr=sr, figure=fig_noisy, ax=ax_noisy)
     enh_im = spec_im(enhanced, sr=sr, figure=fig_enh, ax=ax_enh)
-
     filter = [speech_upl, noisy_wav, enhanced_wav]
     if mic_input is not None and mic_input != "":
         filter.append(mic_input)
     cleanup_tmp(filter)
-
     return noisy_wav, noisy_im, enhanced_wav, enh_im
 
 
+
 def specshow(
     spec,
     ax=None,