A Recognition Method Using Synthesis-based Scoring That Incorporates Direct Relations between Static and Dynamic Feature Vector Time Series

It is well known that hidden Markov models (HMMs) can only exploit the time-dependence in the speech process in a limited way. Parametric trajectory models have been proposed to exploit this time-dependency. However, parametric trajectory modeling methods are unable to take advantage of efficient HMM training and recognition methods. This paper describes a new speech recognition technique that generates a speech trajectory mean using a HMM-based speech synthesis method. This method generates an acoustic trajectory by maximizing the likelihood of the trajectory taking into account the relation between the cepstrum, deltacepstrum, and delta-delta cepstrum. Speaker dependent and speaker independent speech recognition experiments showed that the proposed method is effective for speech recognition.