Fix vc_fn #3

app.py

@@ -15,6 +15,8 @@ logging.getLogger('matplotlib').setLevel(logging.WARNING)
 
 limitation = os.getenv("SYSTEM") == "spaces"  # limit audio length in huggingface spaces
 
+INFERENCE_OUTPUT_DIRNAME = '/output/'
+
 audio_postprocess_ori = gr.Audio.postprocess
 
 def audio_postprocess(self, y):
@@ -44,23 +46,49 @@ def vc_fn(input_audio, vc_transform, voice):
                                    )
     return "Success", (44100, out_audio.cpu().numpy())
 
-def get_speakers():
-  speakers = []
 
-  for _,dirs,_ in os.walk("/models"):
-    for folder in dirs:
-      cur_speaker = {}
-      # Look for G_****.pth
-      g = glob.glob(os.path.join("/models",folder,'G_*.pth'))
-      if not len(g):
-        continue
-      cur_speaker["model_path"] = g[0]
-      cur_speaker["model_folder"] = folder
-      cur_speaker["name"] = folder
+def run_inference(input_audio, speaker):
+    if input_audio is None:
+        return "You need to upload an audio", None
+    sampling_rate, audio = input_audio
+    duration = audio.shape[0] / sampling_rate
+    if duration > 20 and limitation:
+        return "Please upload an audio file that is less than 20 seconds. If you need to generate a longer audio file, please use Colab.", None
+    audio = (audio / np.iinfo(audio.dtype).max).astype(np.float32)
+    if len(audio.shape) > 1:
+        audio = librosa.to_mono(audio.transpose(1, 0))
+    if sampling_rate != 16000:
+        audio = librosa.resample(audio, orig_sr=sampling_rate, target_sr=16000)
+
+	#TODO edit from GUI
+	cluster_ratio = 1
+	noise_scale = 2
+	is_pitch_prediction_enabled = True
+	f0_method = "dio"
+	transpose = 0
+
+    model_path = f"/models/{speaker}/{speaker}.pth"
+    config_path = f"/models/{speaker}/config.json"
+    cluster_path = ""
+
+    raw_path = io.BytesIO()
+    soundfile.write(raw_path, audio, 16000, format="wav")
+    raw_path.seek(0)
+
+    inference_cmd = f"svc infer {raw_path.absolute()} -m {model_path} -c {config_path} {f'-k {cluster_path} -r {cluster_ratio}' if cluster_path != '' and cluster_ratio > 0 else ''} -t {transpose} --f0-method {f0_method} -n {noise_scale} -o {INFERENCE_OUTPUT_DIRNAME}/{raw_path.name} {'' if is_pitch_prediction_enabled else '--no-auto-predict-f0'}"
 
-      speakers.append(copy.copy(cur_speaker))
+    # out_audio, out_sr = model.infer(sid, vc_transform, raw_path,
+    #                              auto_predict_f0=True,
+    #                               )
 
-  return sorted(speakers, key=lambda x:x["name"].lower())
+	result = subprocess.run(
+		inference_cmd.split(),
+		stdout=subprocess.PIPE,
+		stderr=subprocess.STDOUT,
+		text=True
+	  )
+	print(result)
+    return "Success", "TODO" # (44100, out_audio.cpu().numpy())
 
 if __name__ == '__main__':
     parser = argparse.ArgumentParser()
@@ -69,8 +97,7 @@ if __name__ == '__main__':
     parser.add_argument("--share", action="store_true", default=False, help="share gradio app")
     args = parser.parse_args()
 
-    speakers = get_speakers()
-    speaker_list = [x["name"] for x in speakers]
+    speakers = ["chapaev", "petka", "anka"]
 
     models = []
     voices = []
@@ -80,20 +107,16 @@ if __name__ == '__main__':
     with gr.Blocks() as app:
         gr.Markdown(
             "# <center> Sovits Chapay\n"
-            "## <center> The input audio should be clean and pure voice without background music.\n"
         )
 
         with gr.Row():
             with gr.Column():
                 vc_input = gr.Audio(label="Input audio"+' (less than 20 seconds)' if limitation else '')
-
-                vc_transform = gr.Number(label="vc_transform", value=0)
-
-                voice = gr.Dropdown(choices=speaker_list, visible=True)
+                speaker = gr.Dropdown(choices=speakers, visible=True)
 
                 vc_submit = gr.Button("Generate", variant="primary")
             with gr.Column():
                 vc_output1 = gr.Textbox(label="Output Message")
-                vc_output2 = gr.Audio(label="Output Audio")
-            vc_submit.click(vc_fn, [vc_input, vc_transform, voice], [vc_output1, vc_output2])
+                # vc_output2 = gr.Audio(label="Output Audio")
+            vc_submit.click(run_inference, [vc_input, speaker], [vc_output1, vc_output2])
         app.queue(concurrency_count=1, api_open=args.api).launch(share=args.share)