svm_training

Copyright 2014-2019 Anthony Larcher

svm_training provides utilities to train Support Vector Machines to perform speaker verification.

svm_training.svm_training(svmDir, background_sv, enroll_sv, num_thread=1)

Train Suport Vector Machine classifiers for two classes task (as implemented for nowbut miht change in the future to include multi-class classification) Training is parallelized on multiple threads.

Parameters

• svmDir – directory where to store the SVM models

• background_sv – StatServer of super-vectors for background impostors. All super-vectors are used without selection

• enroll_sv – StatServer of super-vectors used for the target models

• num_thread – number of thread to launch in parallel

svm_training.svm_training_singleThread(K, msn, bsn, svm_dir, background_sv, models, enroll_sv)

Train Suport Vector Machine classifiers for two classes task (as implemented for nowbut miht change in the future to include multi-class classification)

Parameters

• K – pre-computed part of the Gram matrix

• msn – maximum number of sessions to train a SVM

• bsn – number of session used as background impostors

• svm_dir – directory where to store the SVM models

• background_sv – StatServer of super-vectors for background impostors. All super-vectors are used without selection

• models – list of models to train. The models must be included in the enroll_sv StatServer

• enroll_sv – StatServer of super-vectors used for the target models