We can extract both vocals and music background from this repo ? about music_source_separation

[qiuqiangkong]

Hi nguyenlm, Thanks for the remind! Now you can follow https://github.com/bytedance/music_source_separation/blob/master/separate_scripts/download_checkpoints.sh to download the resunet_subbandtime checkpoints.

…

On Thu, 16 Sept 2021 at 16:21, nguyenlm ***@***.***> wrote: Hey @qiuqiangkong <https://github.com/qiuqiangkong>, In the separate_scripts/download_checkpoints.sh you downloaded the ismir2021 checkpoint, but in the separate_scripts/separate_vocals.sh the default model was resunet_subbandtime. That leads to mismatch error 😄. I've tried the resunet_ismir2021 model to separate vocals, it reduced about 70% the accompaniment in the audio, that's awesome. I can improve more by finetuning your pretrained model ? Btw you can release resunet_subbandtime model. Again, thanks for your amazing work. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#1 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ADFXTSPUZSWG3HIUWFCJKDTUCGSJJANCNFSM5EAXQSTQ> . Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub>.

[qiuqiangkong]

Yes, both vocals and accompaniment are supported.

[leminhnguyen]

the pretrained model doesn't seem to work. I tried to download the checkpoints from your script, after that run the separate_vocals.sh but the size mismatch problem was raised:

size mismatch for stft.conv_real.weight: copying a param with shape torch.Size([1025, 1, 2048]) from checkpoint, the shape in current model is torch.Size([257, 1, 512]).
size mismatch for stft.conv_imag.weight: copying a param with shape torch.Size([1025, 1, 2048]) from checkpoint, the shape in current model is torch.Size([257, 1, 512]).
size mismatch for istft.ola_window: copying a param with shape torch.Size([2048]) from checkpoint, the shape in current model is torch.Size([512]).
size mismatch for istft.conv_real.weight: copying a param with shape torch.Size([2048, 2048, 1]) from checkpoint, the shape in current model is torch.Size([512, 512, 1]).
size mismatch for istft.conv_imag.weight: copying a param with shape torch.Size([2048, 2048, 1]) from checkpoint, the shape in current model is torch.Size([512, 512, 1]).
size mismatch for bn0.weight: copying a param with shape torch.Size([1025]) from checkpoint, the shape in current model is torch.Size([257]).
size mismatch for bn0.bias: copying a param with shape torch.Size([1025]) from checkpoint, the shape in current model is torch.Size([257]).
size mismatch for bn0.running_mean: copying a param with shape torch.Size([1025]) from checkpoint, the shape in current model is torch.Size([257]).
size mismatch for bn0.running_var: copying a param with shape torch.Size([1025]) from checkpoint, the shape in current model is torch.Size([257]).
size mismatch for encoder_block1.conv_block1.bn1.weight: copying a param with shape torch.Size([2]) from checkpoint, the shape in current model is torch.Size([8]).
size mismatch for encoder_block1.conv_block1.bn1.bias: copying a param with shape torch.Size([2]) from checkpoint, the shape in current model is torch.Size([8]).
size mismatch for encoder_block1.conv_block1.bn1.running_mean: copying a param with shape torch.Size([2]) from checkpoint, the shape in current model is torch.Size([8]).
size mismatch for encoder_block1.conv_block1.bn1.running_var: copying a param with shape torch.Size([2]) from checkpoint, the shape in current model is torch.Size([8]).
size mismatch for encoder_block1.conv_block1.conv1.weight: copying a param with shape torch.Size([32, 2, 3, 3]) from checkpoint, the shape in current model is torch.Size([32, 8, 3, 3]).
size mismatch for encoder_block1.conv_block1.shortcut.weight: copying a param with shape torch.Size([32, 2, 1, 1]) from checkpoint, the shape in current model is torch.Size([32, 8, 1, 1]).
size mismatch for after_conv2.weight: copying a param with shape torch.Size([8, 32, 1, 1]) from checkpoint, the shape in current model is torch.Size([32, 32, 1, 1]).
size mismatch for after_conv2.bias: copying a param with shape torch.Size([8]) from checkpoint, the shape in current model is torch.Size([32]).

[qiuqiangkong]

Did you change the hyper-parameters such as window_size in the model? They should be 2048 by default.

…

On Thu, 16 Sept 2021 at 13:42, nguyenlm ***@***.***> wrote: the pretrained model doesn't seem to work. I tried to download the checkpoints from your script, after that run the separate_vocals.sh but the size mismatch problem was raised: size mismatch for stft.conv_real.weight: copying a param with shape torch.Size([1025, 1, 2048]) from checkpoint, the shape in current model is torch.Size([257, 1, 512]). size mismatch for stft.conv_imag.weight: copying a param with shape torch.Size([1025, 1, 2048]) from checkpoint, the shape in current model is torch.Size([257, 1, 512]). size mismatch for istft.ola_window: copying a param with shape torch.Size([2048]) from checkpoint, the shape in current model is torch.Size([512]). size mismatch for istft.conv_real.weight: copying a param with shape torch.Size([2048, 2048, 1]) from checkpoint, the shape in current model is torch.Size([512, 512, 1]). size mismatch for istft.conv_imag.weight: copying a param with shape torch.Size([2048, 2048, 1]) from checkpoint, the shape in current model is torch.Size([512, 512, 1]). size mismatch for bn0.weight: copying a param with shape torch.Size([1025]) from checkpoint, the shape in current model is torch.Size([257]). size mismatch for bn0.bias: copying a param with shape torch.Size([1025]) from checkpoint, the shape in current model is torch.Size([257]). size mismatch for bn0.running_mean: copying a param with shape torch.Size([1025]) from checkpoint, the shape in current model is torch.Size([257]). size mismatch for bn0.running_var: copying a param with shape torch.Size([1025]) from checkpoint, the shape in current model is torch.Size([257]). size mismatch for encoder_block1.conv_block1.bn1.weight: copying a param with shape torch.Size([2]) from checkpoint, the shape in current model is torch.Size([8]). size mismatch for encoder_block1.conv_block1.bn1.bias: copying a param with shape torch.Size([2]) from checkpoint, the shape in current model is torch.Size([8]). size mismatch for encoder_block1.conv_block1.bn1.running_mean: copying a param with shape torch.Size([2]) from checkpoint, the shape in current model is torch.Size([8]). size mismatch for encoder_block1.conv_block1.bn1.running_var: copying a param with shape torch.Size([2]) from checkpoint, the shape in current model is torch.Size([8]). size mismatch for encoder_block1.conv_block1.conv1.weight: copying a param with shape torch.Size([32, 2, 3, 3]) from checkpoint, the shape in current model is torch.Size([32, 8, 3, 3]). size mismatch for encoder_block1.conv_block1.shortcut.weight: copying a param with shape torch.Size([32, 2, 1, 1]) from checkpoint, the shape in current model is torch.Size([32, 8, 1, 1]). size mismatch for after_conv2.weight: copying a param with shape torch.Size([8, 32, 1, 1]) from checkpoint, the shape in current model is torch.Size([32, 32, 1, 1]). size mismatch for after_conv2.bias: copying a param with shape torch.Size([8]) from checkpoint, the shape in current model is torch.Size([32]). — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#1 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ADFXTSMSB7R246Z4JQKO7W3UCF7TRANCNFSM5EAXQSTQ> . Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub>.

[leminhnguyen]

Hey @qiuqiangkong, In the separate_scripts/download_checkpoints.sh you downloaded the ismir2021 checkpoint, but in the separate_scripts/separate_vocals.sh the default model was resunet_subbandtime. That leads to mismatch error 😄.

I've tried the resunet_ismir2021 model to separate vocals, it reduced about 70% the accompaniment in the audio, that's awesome. I can improve more by finetuning your pretrained model ? Btw you can release resunet_subbandtime model ?

Again, thanks for your amazing work.