Generally speaking, the speaker-dependence of a speech recognition system comes from speaker-dependent speech feature. The variation of vocal tract shape is one of the major source of inter-speaker variations. In this paper, we address several methods of vocal tract length normalization (VTLN) for large vocabulary continuous speech recognition: (1) explore the bilinear warping VTLN in frequency domain; (2) propose a speaker-specific Bark/Mel scale VTLN in Bark/Mel domain; (3) investigate adaptation of the normalization factor. Our experimental results show that the speaker-specific Bark/Mel scale VTLN is better than the piecewise/bilinear warping VTLN in frequency domain. It can reduce up to 12% word error rate for our Spanish and English spontaneous speech scheduling task database. For adaptation of the normalization factor, our experimental results show that promising result can be obtained by using not more than three utterances from a speaker to estimate the normalization factor, and the unsupervised adaptation mode works as well as the supervised one. Therefore, the computational complexity of VTLN can be avoided by learning the normalization factor from very few utterances of a new speaker.

22 pages

Return to: SCS Technical Report Collection
School of Computer Science homepage

This page maintained by reports@cs.cmu.edu