The motivation behind the i-vector is to provide an intermediate representation between :

  • high-dimensional GMM super-vector
  • low-dimensional MFCC features

It models the total variability of both the speaker and the channel.

I-vector feature extraction

I-vectors represent the GMM super-vector using a single total-variability subspace.

Like it? Buy me a coffeeLike it? Buy me a coffee

Leave a comment