Update app.py

app.py

@@ -9,10 +9,8 @@ from modelscope.models.audio.tts import SambertHifigan
 from modelscope.pipelines import pipeline
 from modelscope.utils.constant import Tasks
 
-import torch
-import torchaudio
-from denoiser import pretrained
-from denoiser.dsp import convert_audio
+from voicefixer import VoiceFixer
+voicefixer = VoiceFixer()
 
 # model_0
 
@@ -115,18 +113,12 @@ def extend(audio):
 
 # denoise
 
-model1 = pretrained.dns64().cuda()
-
-def denoise(audio):
-
-    wav, sr = torchaudio.load(audio)
-    wav = convert_audio(wav.cuda(), sr, model1.sample_rate, model1.chin)
-    with torch.no_grad():
-        denoised = model1(wav[None])[0]
-
-    write("denoised.wav", model1.sample_rate, denoised.data.cpu().numpy())
-    
-    return "denoised.wav"
+def inference(audio):
+    voicefixer.restore(input=audio.name, # input wav file path
+                    output="output.wav", # output wav file path
+                    cuda=False, # whether to use gpu acceleration
+                    mode = int(0)) # You can try out mode 0, 1 to find out the best result
+    return 'output.wav'
 
 
 app = gr.Blocks()
@@ -151,7 +143,7 @@ with app:
         btn.click(fn=infer, inputs=[inp], outputs=[out])
         btn1.click(fn=infer1, inputs=[inp], outputs=[out])
         btn2.click(fn=extend, inputs=[out], outputs=[out1])
-        btn2.click(fn=denoise, inputs=[out1], outputs=[out2])
+        btn2.click(fn=inference, inputs=[out1], outputs=[out2])
 
     gr.Markdown("### <center>注意❗：请不要生成会对个人以及组织造成侵害的内容，此程序仅供科研、学习及个人娱乐使用。</center>")
     gr.HTML('''