Update app.py

app.py

@@ -82,6 +82,31 @@ def infer1(text):
         f.write(output["output_wav"])
     return filename + "file.wav"
 
+# upsample
+
+import numpy as np
+import torch
+from hifi_gan_bwe import BandwidthExtender
+
+MAX_LENGTH = 600.0
+
+model = BandwidthExtender.from_pretrained("hifi-gan-bwe-10-42890e3-vctk-48kHz")
+
+def extend(audio):
+    fs, x = audio
+    x = x[:int(MAX_LENGTH * fs)]
+    x = x.astype(np.float32) / 32767.0
+    if len(x.shape) == 1:
+        x = x[:, np.newaxis]
+
+    with torch.no_grad():
+        y = np.stack([model(torch.from_numpy(x), fs) for x in x.T]).T
+        y = (y * 32767.0).astype(np.int16)
+        fs = int(model.sample_rate)
+
+    return fs, y
+
+
 app = gr.Blocks()
 
 with app:
@@ -97,9 +122,14 @@ with app:
             with gr.Row():
                 btn = gr.Button("使用AI娜娜的声音", variant="primary")
                 btn1 = gr.Button("使用AI小杰的声音", variant="primary")
-        out = gr.Audio(label="为您生成的专属音频", type="filepath")
-
+        with gr.Column():
+            with gr.Row():
+                out = gr.Audio(label="为您生成的专属音频", type="filepath")
+                out1 = gr.Audio(label="更高采样率的专属音频", type="filepath")
+            btn2 = gr.Button("一键提高采样率")
+    
         btn.click(fn=infer, inputs=[inp], outputs=[out])
         btn1.click(fn=infer1, inputs=[inp], outputs=[out])
+        btn2.click(fn=extend, inputs=[out], outputs=[out1])
         
 app.launch(show_error=True)