yeyupiaoling / voiceprintrecognition-paddlepaddle Goto Github PK

大佬我执行 python3 infer_contrast.py --audio_path1=audio/a_1.wav --audio_path2=audio/b_2.wav 的时候报 ValueError: The path (models/ecapa_tdnn/model.pdparams) to load model not exists. 同时我没有安装PaddlePaddle的Inference预测库，因为我在安装的时候总是报不适用当前的平台，但是其他环境我都安装完了，是因为这个的原因吗？顺便PaddlePaddle的Inference预测库只能是Nvidia Jetson吗？

想通过对采集的音频进行去静音处理提高识别率，请问大佬有什么好的方法没？

如题 ,感谢!

将中文语音语料数据集下载好后，运行create_data.py发现数据读取非常的慢

which configs need I to modify

Hi,

Great job!

I encountered a problem when using it. When there is only one data in the audio_db(just resigter one wav), an error will occur. Do you have any suggestions

请选择功能，0为注册音频到声纹库，1为执行声纹识别：1
Traceback (most recent call last):
File "/usr/local/src/VoiceprintRecognition-PaddlePaddle/infer_recognition.py", line 51, in
name = predictor.recognition(audio_data)
File "/usr/local/src/VoiceprintRecognition-PaddlePaddle/ppvector/predict.py", line 300, in recognition
name = self.__retrieval(np_feature=[feature])[0]
File "/usr/local/src/VoiceprintRecognition-PaddlePaddle/ppvector/predict.py", line 160, in __retrieval
similarity = cosine_similarity(self.audio_feature, feature.reshape(1, -1)).squeeze()
File "/data1/bigdata/miniconda3/envs/paddlespeech/lib/python3.9/site-packages/sklearn/metrics/pairwise.py", line 1393, in cosine_similarity
X, Y = check_pairwise_arrays(X, Y)
File "/data1/bigdata/miniconda3/envs/paddlespeech/lib/python3.9/site-packages/sklearn/metrics/pairwise.py", line 155, in check_pairwise_arrays
X = check_array(
File "/data1/bigdata/miniconda3/envs/paddlespeech/lib/python3.9/site-packages/sklearn/utils/validation.py", line 902, in check_array
raise ValueError(
ValueError: Expected 2D array, got 1D array instead:
array=[ -9.528404 20.831924 -7.2488213 -10.465841 3.648481
-27.02867 -21.46711 -7.2084546 -2.5436175 16.762877
-23.119123 -19.69215 19.306414 -29.709349 10.007728
9.711845 15.237292 -31.4283 13.326845 24.98609
-0.93558085 -13.328272 -1.4318435 -2.5589817 -14.899953
-10.004118 8.370364 -28.427952 -16.635942 17.125128
-21.187462 -13.563347 22.93637 0.2699321 -42.41188
-32.501728 -0.88023186 -17.82453 22.414608 -4.979337
-8.525855 23.49937 2.4326832 45.067253 -23.60708
-30.05538 1.6135166 -40.467884 6.419506 -22.83227
14.336002 -6.9231305 -2.9081142 -3.9221501 22.34546
15.799733 -31.135666 -11.1763735 -36.390778 20.186132
-1.3171989 3.5721273 -6.8223796 -0.87155807 0.6096292
-0.7906767 24.010586 -30.601904 42.77444 4.056578
8.387045 -32.088486 1.4989619 -16.874323 -14.909355
10.0754 26.727545 3.1605248 -7.187451 17.319191
-29.09326 -10.794649 16.416176 -39.16383 8.402718
-18.068346 -24.327047 -10.149664 32.352417 29.281029
23.015427 39.01204 -0.17142385 18.62544 -43.314125
-3.381555 -24.813742 -9.385822 -14.046436 -30.988276
-31.660748 -11.767179 12.45567 -14.2988205 -18.520817
-4.230826 7.557493 19.553474 30.747616 -22.761354
9.038652 30.985561 -43.875 -6.442013 -13.843025
-22.249443 -9.539998 -8.984698 -6.8686695 -10.849445
-3.2795053 -10.321569 20.782562 -33.36194 29.603746
5.8829403 -16.764643 33.195232 26.174906 -10.609048
-0.93014306 21.179016 3.711417 -30.199566 -3.0830624
-15.195986 3.977836 -22.489988 -32.214226 -38.9151
11.913775 5.5171957 -16.352848 -14.75191 -25.292871
12.144376 9.523496 31.895811 12.43035 25.481085
19.512949 2.1031296 28.158194 0.76135576 6.5883613
2.3754144 -9.671471 26.664793 15.48469 -17.29369
6.898872 11.445733 9.200221 -12.630247 -10.417504
0.6524699 -0.5872514 -9.257348 -18.316444 -13.710389
7.0560064 -28.609156 8.623157 -5.470796 -3.8638942
16.686884 -11.8199415 -30.546083 16.687752 9.732939
-27.76715 -24.231022 -3.1489227 -10.5073 -0.15581396
7.884112 -11.454353 -6.92933 20.55611 -7.311349
25.531921 -8.121537 ].
Reshape your data either using array.reshape(-1, 1) if your data has a single feature or array.reshape(1, -1) if it contains a single sample.

老哥，你试过你新改的这个代码转onnx吗？好像不成功
File "/data1/sda/lingck/VoiceRecogntion/VoiceprintRecognition-PaddlePaddle/modules/ecapa_tdnn.py", line 186, in forward
y_i = self.blocksi - 1
else:
y_i = self.blocks[i - 1](x_i + y_i)
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ <--- HERE
y.append(y_i)
y = paddle.concat(y, axis=1)

UnboundLocalError: local variable 'y_i' referenced before assignment

请问各位大佬，我这边有几百条标记好讲话人的语音作为训练集，打算之后用训练好的模型去给更多的语音标记讲话人。

那我是直接拿我的这几百条去训练和使用，还是先用zhvoice 的数据训练完，再用我自己这几百条去训练呢？