Update app.py

app.py

@@ -153,20 +153,20 @@ def spectrogram_to_audio(magnitude_spectrogram):
     # Inverse STFT to convert the spectrogram back to audio
     audio = torch.istft(complex_spectrogram, n_fft=n_fft, hop_length=hop_length)
 
-    # Check for NaNs or Infs in the audio and replace them with zeros
+    # Handle any NaNs or Infs
     audio = torch.nan_to_num(audio, nan=0.0, posinf=0.0, neginf=0.0)
 
-    # Normalize audio to the range [-1, 1]
+    # Normalize audio to the range [-1, 1] (standard audio range)
     if torch.max(torch.abs(audio)) != 0:
         audio = audio / torch.max(torch.abs(audio))
 
-    # Clip the audio to ensure it fits in the range [-1, 1]
+    # Clip the audio to the range [-1, 1] to avoid out-of-bounds values
     audio = torch.clamp(audio, min=-1, max=1)
 
-    # Convert to 16-bit PCM format by scaling and casting to int16
+    # Convert the audio to 16-bit PCM format by scaling and casting to int16
     audio = (audio * 32767).short()
 
-    # Ensure the values are clipped to the int16 range
+    # Ensure audio is in the valid range for int16
     audio = torch.clamp(audio, min=-32768, max=32767)
 
     return audio
@@ -183,7 +183,7 @@ def generate_audio_from_image(image):
     generated_audio = spectrogram_to_audio(generated_spectrogram.squeeze(0).cpu())
 
     # Convert the audio tensor to a NumPy array before passing it to Gradio
-    generated_audio_numpy = generated_audio.numpy()
+    generated_audio_numpy = generated_audio.numpy().astype(np.int16)  # Ensure it's int16
 
     # Return the sample rate and the NumPy array containing the audio
     return (sample_rate, generated_audio_numpy)