update

app.py

@@ -31,7 +31,7 @@ def mix_at_snr(clean, noise, snr, eps=1e-10):
     if noise.shape[1] < clean.shape[1]:
         noise = noise.repeat((1, int(math.ceil(clean.shape[1] / noise.shape[1]))))
     max_start = int(noise.shape[1] - clean.shape[1])
-    start = torch.randint(0, max_start)
+    start = torch.randint(0, max_start, ()).item()
     noise = noise[:, start : start + clean.shape[1]]
     E_speech = torch.mean(clean.pow(2)) + eps
     E_noise = torch.mean(noise.pow(2))
@@ -47,9 +47,13 @@ def as_gradio_audio(x):
     return sr, (x / 0x7FFF).to(torch.int16).cpu().numpy()
 
 
-def mix_and_denoise(speech, noise, snr):
+def mix_and_denoise(speech, speech_alt, noise, snr):
     if noise is None:
         noise = "samples/dkitchen.wav"
+    if speech is None:
+        if speech_alt is None:
+            speech = "samples/p232_013_clean.wav"
+        speech = speech_alt
     print(speech, noise, snr)
     sr = config("sr", 48000, int, section="df")
     speech, _ = load_audio(speech, sr)
@@ -64,10 +68,13 @@ def mix_and_denoise(speech, noise, snr):
 
 inputs = [
     gradio.inputs.Audio(
-        source="microphone,upload", type="filepath", optional=True, label="Speech"
+        source="microphone", type="filepath", optional=True, label="Record your own voice"
     ),
-    gradio.inputs.Audio(source="upload", type="filepath", optional=True, label="Noise"),
-    gradio.inputs.Slider(minimum=-10, maximum=40, step=5, default=10),
+    gradio.inputs.Audio(
+        source="upload", type="filepath", optional=True, label="Alternative: Upload speech sample"
+    ),
+    gradio.inputs.Audio(source="upload", type="filepath", optional=True, label="Upload noise sample"),
+    gradio.inputs.Slider(minimum=-20, maximum=40, step=5, default=10),
 ]
 examples = [
     ["samples/p232_013_clean.wav", "samples/dkitchen.wav", 10],