An acoustic model adaptation using HMM-based speech synthesis

Recently, personal digital assistants like cellular phones are shifting to the IP terminal. The encoding-decoding process utilized for transmitting over IP networks deteriorates the quality of the speech data. This deterioration causes degradation in speech recognition performance. Acoustic model adaptations could improve recognition performance. However, the current adaptation methods usually require a large amount of adaptation data. A novel adaptation method using speech synthesis based on HMM (hidden Markov model) is proposed. This method does not require speech data for adaptation because speech data is generated by speech synthesis from the acoustic model. Experimental results on G.723.1 coded speech recognition show that the proposed method improves speech recognition performance. A relative improvement in word accuracy of approximately 2% was observed.