Improved Syllable Based Acoustic Modeling by Inter-Syllable Transition Model for Continuous Chinese Speech Recognition

Accurately modeling the acoustic variabilities caused by coarticulation is important in continuous speech recognition. Recent research indicates that syllable units do better in modeling intra-syllable co-articulation effect than sub-syllable units. However, most continuous Mandarin speech recognition systems use context dependent phones or initial/finals (IFs) as the basic acoustic unit because it is difficult to collect sufficient data to train longer units. Here we present a syllable based approach which includes two steps. Firstly, context independent syllable based acoustic models are trained, and the models are initialized by intra-syllable IFs based diphones to solve the problem of training data sparsity. Secondly, we capture the inter-syllable co-articulation effect by incorporating inter-syllable transition models into the recognition system. Experiment results show that the acoustic model based on the presented approach is effective in improving the recognition performance.