Update app.py

app.py

@@ -1,4 +1,4 @@
-import numpy as numpy
+import numpy as np
 
 import streamlit as st
 import librosa
@@ -32,8 +32,8 @@ def load_model():
     return session, onnx_model, input_names, output_names
 
 def inference(re_im, session, onnx_model, input_names, output_names):
-    inputs = {input_names[i]: numpy.zeros([d.dim_value for d in _input.type.tensor_type.shape.dim],
-                                       dtype=numpy.float32)
+    inputs = {input_names[i]: np.zeros([d.dim_value for d in _input.type.tensor_type.shape.dim],
+                                       dtype=np.float32)
               for i, _input in enumerate(onnx_model.graph.input)
               }
 
@@ -46,25 +46,25 @@ def inference(re_im, session, onnx_model, input_names, output_names):
         inputs[input_names[3]] = mlp_state
         output_audio.append(out)
 
-    output_audio = torch.tensor(numpy.concatenate(output_audio, 0))
+    output_audio = torch.tensor(np.concatenate(output_audio, 0))
     output_audio = output_audio.permute(1, 0, 2).contiguous()
     output_audio = torch.view_as_complex(output_audio)
     output_audio = torch.istft(output_audio, window, stride, window=hann)
-    return output_audio.numpy()
+    return output_audio.np()
 
 def visualize(hr, lr, recon, sr):
     sr = sr
     window_size = 1024
-    window = numpy.hanning(window_size)
+    window = np.hanning(window_size)
 
     stft_hr = librosa.core.spectrum.stft(hr, n_fft=window_size, hop_length=512, window=window)
-    stft_hr = 2 * numpy.abs(stft_hr) / numpy.sum(window)
+    stft_hr = 2 * np.abs(stft_hr) / np.sum(window)
 
     stft_lr = librosa.core.spectrum.stft(lr, n_fft=window_size, hop_length=512, window=window)
-    stft_lr = 2 * numpy.abs(stft_lr) / numpy.sum(window)
+    stft_lr = 2 * np.abs(stft_lr) / np.sum(window)
 
     stft_recon = librosa.core.spectrum.stft(recon, n_fft=window_size, hop_length=512, window=window)
-    stft_recon = 2 * numpy.abs(stft_recon) / numpy.sum(window)
+    stft_recon = 2 * np.abs(stft_recon) / np.sum(window)
 
     fig, (ax1, ax2, ax3) = plt.subplots(3, 1, sharey=True, sharex=True, figsize=(16, 12))
     ax1.title.set_text('Оригинальный сигнал')
@@ -109,7 +109,7 @@ lossy_input = lossy_input.reshape(-1)
 hann = torch.sqrt(torch.hann_window(window))
 lossy_input_tensor = torch.tensor(lossy_input)
 re_im = torch.stft(lossy_input_tensor, window, stride, window=hann, return_complex=False).permute(1, 0, 2).unsqueeze(
-    1).numpy().astype(np.float32)
+    1).np().astype(np.float32)
 session, onnx_model, input_names, output_names = load_model()
 
 if st.button('Сгенерировать потери'):