search

alex-berard / seq2seq

Attention-based sequence to sequence learning
Apache License 2.0
388 stars 122 forks source link

MFCC EXTRACT #26

Open DonnieZhang586 opened 5 years ago

DonnieZhang586 commented 5 years ago

HI,sir！ Your open source code is very exciting, its features are powerful，I am currently doing end-to-end speech translation tasks（chinese speech to english text），I want to refer to your code.I have a few questions that would like to ask you.1、extract.py and extract-new.py What is the difference between these two codes?What are their respective functions?

alex-berard commented 5 years ago

Hi, extract.py was used to compute MFCCs in this paper and this paper. However, it is hard to setup because it relies on the Yaafe library, which is kind of deprecated...

extract-new.py should be easier to install, because it relies on scipy.io.wavfile, which you can install with a simple pip command. It also computes log-mel filterbank features by default (rather than MFCCs), which seems to be much more common and might work better. However, I haven't really tested it.

DonnieZhang586 commented 5 years ago

Hi, extract.py was used to compute MFCCs in this paper and this paper. However, it is hard to setup because it relies on the Yaafe library, which is kind of deprecated...

extract-new.py should be easier to install, because it relies on scipy.io.wavfile, which you can install with a simple pip command. It also computes log-mel filterbank features by default (rather than MFCCs), which seems to be much more common and might work better. However, I haven't really tested it.

thank you for replying to my question during your busy schedule,I am currently doing end-to-end speech translation tasks（chinese speech to english text）,i found two vocabularies used in your code and your paper ,do you need to used the source language text and the target language text in the chinese to english task?