Restructuring HMM states for speaker adaptation in Mandarin speech recognition

With the tendency of posterior probability taken into account, a state-restructuring method is proposed based on confusions between HMM states. In the method, HMM state is restructured by sharing Gaussian components with its related states and the re-estimation of the increased-parameters, i.e., the inter-state weights, is derived under the EM framework. Experiments are performed on speaker-independent large vocabulary continuous Mandarin speech recognition. The results show the state-restructured systems outperform the baseline system and the combining with MLLR adaptation can lead to consistent and significant improvement on recognition accuracy over MLLR.