Universal Background Model
A large GMM (hundreds to thousands of components) trained on pooled speech from many speakers models 'speech in general'; a target speaker's model is MAP-adapted from it, and verification scores the likelihood ratio between the speaker model and the UBM.
This Concept is waiting for its first lesson!
A large GMM (hundreds to thousands of components) trained on pooled speech from many speakers models 'speech in general'; a target speaker's model is MAP-adapted from it, and verification scores the likelihood ratio between the speaker model and the UBM.
Are you a teacher? Sign in to start contributing.
Sign In