search

JusperLee / Dual-Path-RNN-Pytorch

Dual-path RNN: efficient long sequence modeling for time-domain single-channel speech separation implemented by Pytorch
Apache License 2.0
398 stars 68 forks source link

关于音频读入的问题 #60

Open FBSalol opened 1 month ago

FBSalol commented 1 month ago

作者你好，我在加载音频时，读入的src参数维数不是二维，此问题在加载不同音频时时有时无，请问作者遇到过类似的问题吗，是如何解决的呢？谢谢 280b90e5d4ef5ff095301789b3889bb Traceback (most recent call last): File "D:\model\Dual-Path-RNN-Pytorch-master\Dual-Path-RNN-Pytorch-master\dualrnn_test_wav.py", line 73, in main() File "D:\model\Dual-Path-RNN-Pytorch-master\Dual-Path-RNN-Pytorch-master\dualrnn_test_wav.py", line 69, in main separation.inference(args.save_path) File "D:\model\Dual-Path-RNN-Pytorch-master\Dual-Path-RNN-Pytorch-master\dualrnn_test_wav.py", line 50, in inference write_wav(filename, s, 16000) File "D:\model\Dual-Path-RNN-Pytorch-master\Dual-Path-RNN-Pytorch-master\data_loader\AudioReader.py", line 37, in write_wav torchaudio.save(fname, src, sample_rate) File "C:\Users\dell\AppData\Local\conda\conda\envs\d2l\lib\site-packages\torchaudio_backend\utils.py", line 288, in save return backend.save(uri, src, sample_rate, channels_first, format, encoding, bits_per_sample, buffer_size) File "C:\Users\dell\AppData\Local\conda\conda\envs\d2l\lib\site-packages\torchaudio_backend\soundfile.py", line 39, in save soundfile_backend.save( File "C:\Users\dell\AppData\Local\conda\conda\envs\d2l\lib\site-packages\torchaudio_backend\soundfile_backend.py", line 427, in save raise ValueError(f"Expected 2D Tensor, got {src.ndim}D.") ValueError: Expected 2D Tensor, got 1D.

JusperLee commented 1 month ago

你可以用view(B, 2, T)

heyun913 commented 1 month ago

hello ，解决了么，遇到了同样的问题

FBSalol commented 1 month ago

hello ，解决了么，遇到了同样的问题

没有，估计大概率是音频的原因，我使用自己录制的音频就不会出现此问题，你可以看看作者提供的混合音频的代码