Knowledge-Rich Model Transformations for Speaker Normalization in Speech Recognition

In this work we extend the test utterance adaptation technique used in vocal tract length normalization to a larger number of speaker characteristic features. We perform partially joint estimation of four features: the VTLN warping factor, the corner position of the piece-wise linear warping function, spectral tilt in voiced segments, and model variance scaling. In experiments on the Swedish PF-Star children database, joint estimation of warping factor and variance scaling lowers the recognition error rate compared to warping factor alone.