search

mravanelli / SincNet

SincNet is a neural architecture for efficiently processing raw audio samples.
MIT License
1.14k stars 263 forks source link

Amplification factor #81

Closed CarmiShimon closed 4 years ago

CarmiShimon commented 4 years ago

Hi and thanks again for replying the issues and your great work! I wonder: in create_batches_rnd function - rand_amp_arr could be greater then 1 after the multiplication so it could make the audio 'clipping'. What do you think about it?

Thanks a lot, Carmi

mravanelli commented 4 years ago

Hi, thank you for the question. Actually, in the config file we use " cnn_use_laynorm_inp=True" that normalizes every time the dynamic and make rand_amp useless. Without layer norm set to "true", you might have signals with amplitude

  1. This is not real clipping (we are not saving the signal), but it could be something that one wants to avoid.

Best,

Mirco

On Wed, 18 Dec 2019 at 10:21, carmi shimon notifications@github.com wrote:

Hi and thanks again for replying the issues and your great work! I wonder: in create_batches_rnd function - rand_amp_arr could be greater then 1 after the multiplication so it could make the audio 'clipping'. What do you think about it?

Thanks a lot, Carmi

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/mravanelli/SincNet/issues/81?email_source=notifications&email_token=AEA2ZVUJUHMMVMVWXG7DF6TQZI5YNA5CNFSM4J4MVMR2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4IBL43LQ, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEA2ZVWPNDHVQTI6B4T4SZ3QZI5YNANCNFSM4J4MVMRQ .

CarmiShimon commented 4 years ago

Thanks a lot for the detailed explanation! Carmi