13th Annual Conference of the International Speech Communication Association

Portland, OR, USA
September 9-13, 2012

IVN-Based Joint Training Of GMM And HMMs Using An Improved VTS-Based Feature Compensation For Noisy Speech Recognition

Jun Du, Qiang Huo

Microsoft Research Asia, Beijing, China

In our previous work, we proposed a feature compensation approach using high-order vector Taylor series approximation for noisy speech recognition. In this paper, first we improve the feature compensation in both efficiency and accuracy by boosted mixture learning of GMM, applying higher order information of VTS approximation only to the noisy speech mean parameters, acoustic context expansion, and modeling the convolutional distortion as a single Gaussian. Then we design a procedure to perform irrelevant variability normalization based joint training of GMM and HMM using VTS-based feature compensation. The effectiveness of our proposed approach is confirmed by experiments on Aurora3 database.

Index Terms: irrelevant variability normalization, feature compensation, vector Taylor series.

Full Paper

Bibliographic reference.  Du, Jun / Huo, Qiang (2012): "IVN-based joint training of GMM and HMMs using an improved VTS-based feature compensation for noisy speech recognition", In INTERSPEECH-2012, 1227-1230.