search

nouiz / lisa_emotiw

Challenge de reco d'émotions sur les visages.
34 stars 24 forks source link

Simple feature extraction (MFCC) from speech on current data #6

Closed yoshua closed 11 years ago

yoshua commented 11 years ago

Use existing libraries for extracting spectral features from audio and compute these features on the challenge data.

dwf commented 11 years ago

George Dahl's preprocessing for speech data is log mel-scale filterbanks, basically the MFCCs before the cepstral transform, on 25ms windows shifted by either 5ms or 10ms. It would be good to have this preprocessing handy as well.

boulanni commented 11 years ago

Pretty much done, see https://github.com/nouiz/lisa_emotiw/wiki/06.-Emotion-Classification-from-Speech

yoshua commented 11 years ago

Please close the ticket when done.