ACCENT ADAPTATION USING SUBSPACE GAUSSIAN MIXTURE MODELS

This paper investigates employment of Subspace Gaussian Mixture Models (SGMMs) for acoustic model adaptation to-wards different accents for English speech recognition. The SGMMs comprise globally-shared and state-specific param-eters which can efficiently be employed for various kinds of acoustic parameter tying. Research results indicate that well-defined sharing of acoustic model parameters in SGMMs can significantly outperform adapted systems based on conven-tional HMM/GMMs. Furthermore, SGMMs rapidly achieve target acoustic models with small amounts of data. Exper-iments performed with US and UK English versions of the Wall Street Journal (WSJ) corpora indicate that SGMMs lead to approximately 20 % and 8 % relative improvements with respect to speaker-independent and speaker-adapted acous-tic models respectively over conventional HMM/GMMs. Finally, we demonstrate that SGMMs adapted only with 1.5 hours can reach performance of HMM/GMMs trained with 18 hours