Update inference.py

inference.py

@@ -27,9 +27,9 @@ def _calculate_f0(input: np.ndarray,length,sr,f0min,f0max,
         return f0.reshape(-1)
 
 
-def get_text(file,transform=1.0):
+def get_text(wav,sr,transform=1.0):
 
-    wav, sr = librosa.load(file,sr=None)
+    #wav, sr = librosa.load(file,sr=None)
     if sr<16000:
         return 'sample rate too low'
     if len(wav.shape) > 1:
@@ -54,15 +54,21 @@ def get_text(file,transform=1.0):
 def getkey(key):
     return np.power(2,key/12.0)
 
-def infer(f,o,speaker,key,reqf0=False):
-    x,sourcef0 = get_text(f,getkey(key))
+def infer(f,speaker,key,reqf0=False):
+    if f is None:
+        return "You need to upload an audio", None
+    sr, audio = f
+    duration = audio.shape[0] / sr
+    if duration > 120:
+        return "请上传小于2min的音频", None
+    audio = (audio / np.iinfo(audio.dtype).max).astype(np.float32)
+    x,sourcef0 = get_text(audio,sr,getkey(key))
     x_lengths = [np.size(x,1)]
     sid = [speaker]
     ort_inputs = {'x':x,'x_lengths':x_lengths,'sid':sid,"noise_scale":[0.667],"length_scale":[1.0],"noise_scale_w":[0.8]} 
     infersession = onnxruntime.InferenceSession("onnxmodel334.onnx")#,providers=['CUDAExecutionProvider'])
     ort_output = infersession.run(['audio'], ort_inputs)
-    sf.write(o,ort_output[0][0][0],22050,'PCM_16',format='wav')
-    o.seek(0,0)
+    #sf.write(o,ort_output[0][0][0],22050,'PCM_16',format='wav')
     genf0=np.array([])
     if reqf0:
         wav, sr = librosa.load(o,sr=None)
@@ -70,5 +76,4 @@ def infer(f,o,speaker,key,reqf0=False):
             f0min=librosa.note_to_hz('C2'),
             f0max=librosa.note_to_hz('C7'))
         genf0=resize2d(genf0,x_lengths[0])
-    o.seek(0,0)
-    return sourcef0.tolist(),genf0.tolist()
+    return 'success',ort_output[0][0][0]#sourcef0.tolist(),genf0.tolist()