This repo to support the publication: Whisper to neutral mapping using cosine similarity maximization in i-vector space for speaker verification
Main components:
-
feature to i-vector as tensorflow layer: i_vec_layer.py: file containing the layer definition. it can be used as follows:
ivec_layer=i_vec_layer(ivec_dim, fdmin, nmix,root+'ubm.gmm',root+'tvmat.h5'); -
cosine similarity objective function. look at train.py for example usage and the objective function.