Combining eigenvoice speaker modeling and VTS-based environment compensation for robust speech recognition

Abstract

1Eigenvoice and vector Taylor series (VTS) are good models for speaker differences and environmental variations separately. However, speaker and environmental variation always coexist in real-world speech. In this paper, we propose to combine eigenvoice and VTS. Specifically, we introduce eigenvoice speaker modeling for the clean speech into VTS's nonlinear mismatch function. In contrast, the standard VTS uses speakerindependent modeling to represent the clean speech, regardless of speaker differences. The eigenvoice coefficients and the noise model parameters are jointly estimated in the new approach. Experimental results on the Aurora2 task show the improved performances of combining eigenvoice and VTS and demonstrate its ability for speaker and noise factorization.

An explicit independence constraint for factorised adaptation in speech recognition

“…(12). 3 The environment transforms are updated while keeping all the other parameters fixed. 4 The speaker clusters are updated while keeping all the other parameter fixed.…”

“…This requires adaptation data for all possible operating conditions. An alternative approach, acoustic factorisation first proposed in 2001 [1], has been adopted by a number of sites very recently e.g., [2,3,4,5,6]. In parallel with the factorisation approach in speech recognition, there is also work along this line in speech synthesis, e.g., [7,8], where the goal is to synthesis the effect of multiple factors, such as speaker, language and emotion.…”