How does the SepFormer handle arbitrary input lengths?

[hananbo26]

The SepFormer, introduced in the paper ATTENTION IS ALL YOU NEED IN SPEECH SEPARATION, is a popular speech separation network implemented in speechbrain.

However, having gone through the paper, I still need help understanding how it works. How does it deal with inputs of arbitrary lengths? Is there a maximal input length? And what happens if this maximal length is exceeded?

[mravanelli]

Hi, you can take a look at the code and the training recipes available in speechbrain here: - https://github.com/speechbrain/speechbrain/blob/develop/recipes/WSJ0Mix/separation/hparams/sepformer.yaml - https://github.com/speechbrain/speechbrain/blob/develop/speechbrain/lobes/models/dual_path.py

…

On Tue, Oct 31, 2023 at 8:00 AM hananbo26 ***@***.***> wrote: The SepFormer, introduced in the paper ATTENTION IS ALL YOU NEED IN SPEECH SEPARATION <https://arxiv.org/pdf/2010.13154.pdf>, is a popular speech separation network implemented in speechbrain. However, having gone through the paper, I still need help understanding how it works. How does it deal with inputs of arbitrary lengths? Is there a maximal input length? And what happens if this maximal length is exceeded? — Reply to this email directly, view it on GitHub <#2226>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AEA2ZVQNCW47PQ72RHOB2VTYCDR6VAVCNFSM6AAAAAA6XSB5XWVHI2DSMVQWIX3LMV43ERDJONRXK43TNFXW4OZVG44TONRQGE> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>