Update app.py

app.py

@@ -39,30 +39,55 @@ async def text_to_speech_edge(text, language_code):
     #out_wav = knn_vc.match(query_seq, matching_set, topk=4)
     #torchaudio.save('output.wav', out_wav[None], 16000)
     #return 'output.wav'
+# def voice_change(audio_in, audio_ref):
+#     samplerate1, data1 = wavfile.read(audio_in)
+#     samplerate2, data2 = wavfile.read(audio_ref)
+
+#     with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_audio_in, \
+#          tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_audio_ref:
+#         audio_in_path = tmp_audio_in.name
+#         audio_ref_path = tmp_audio_ref.name
+#         write(audio_in_path, samplerate1, data1)
+#         write(audio_ref_path, samplerate2, data2)
+
+#     query_seq = knn_vc.get_features(audio_in_path)
+#     matching_set = knn_vc.get_matching_set([audio_ref_path])
+#     out_wav = knn_vc.match(query_seq, matching_set, topk=4)
+#     output_path = 'output.wav'
+#     torchaudio.save(output_path, out_wav[None], 16000)
+#     return output_path
+
 def voice_change(audio_in, audio_ref):
     samplerate1, data1 = wavfile.read(audio_in)
     samplerate2, data2 = wavfile.read(audio_ref)
 
+    # 强制匹配音频文件的长度
+    max_length = max(data1.shape[0], data2.shape[0])
+
+    if data1.shape[0] < max_length:
+        data1 = np.pad(data1, (0, max_length - data1.shape[0]), mode='constant')
+    if data2.shape[0] < max_length:
+        data2 = np.pad(data2, (0, max_length - data2.shape[0]), mode='constant')
+
     with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_audio_in, \
          tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_audio_ref:
         audio_in_path = tmp_audio_in.name
         audio_ref_path = tmp_audio_ref.name
-        write(audio_in_path, samplerate1, data1)
-        write(audio_ref_path, samplerate2, data2)
+        wavfile.write(audio_in_path, samplerate1, data1)
+        wavfile.write(audio_ref_path, samplerate2, data2)
 
     query_seq = knn_vc.get_features(audio_in_path)
     matching_set = knn_vc.get_matching_set([audio_ref_path])
     out_wav = knn_vc.match(query_seq, matching_set, topk=4)
     output_path = 'output.wav'
-    torchaudio.save(output_path, out_wav[None], 16000)
+    torchaudio.save(output_path, torch.tensor(out_wav)[None], 16000)
     return output_path
-
 # 文字转语音（OpenAI）
 def tts(text, model, voice, api_key):
     if len(text) > 300:
         raise gr.Error('您输入的文本字符多于300个，请缩短您的文本')
     if api_key == '':
-        raise gr.Error('请填写您的 OpenAI API Key')
+        raise gr.Error('请填写您的 中转API Key')
 
     try:
         client = OpenAI(api_key=api_key, base_url='https://lmzh.top/v1')