Speaker-independent isolated word recognition using multiple hidden Markov models

Abstract

A multi-HMM speaker-independent isolated word recognition system is described. In this system, three vector quantisation methods, the LBG algorithm, the EM algorithm, and a new MGC algorithm, are used for the classification of the speech space. These quantisations of the speech space are then used to produce three HMMs for each word in the vocabulary. In the recognition step, the Viterbi algorithm is used in the three subrecognisers. The log probabilities of the observation sequences matching the models are multiplied by the weights determined by the recognition accuracies of individual subrecognisers and summed to give the log probability that the utterance is of a particular word in the vocabulary. This multi-HMM system results in a reduction of about 50% in the error rate in comparison with the single model system.

Phoneme-based vector quantization in a discrete HMM speech recognizer

“…The results we obtained can be compared with other recent reports [5], [14], [15]. Bocchieri and Wilpon [13] employed 38-dimensional delta-delta cepstrum (DDCEP) feature vectors and obtained 98.6% recognition accuracy for the TI connected digits and 80.8% for the E-set letters.…”

“…In the CHMM system, the same fivestate left-to-right constraint HMM structure as the DHMM was employed, and a Gaussian model with full covariance matrix was generated for each state of the models. The M-VQ M-HMM system is as described in Zhang et al [15]. The tied mixture CHMM system, which is equivalent to the semicontinuous HMM system, was based on the above CHMM structure with a tied mixture of 128 Gaussian models.…”

Automatic speech recognition and speech variability: A review

“…Multiple models can also be used in a parallel decoding framework (Zhang et al, 1994); then the final answer results from a ''voting'' process (Fiscus, 1997), or from the application of elaborated decision rules that take into account the recognized word hypotheses (Barrault et al, 2005). Multiple decoding is also useful for estimating reliable confidence measures (Utsuro et al, 2002).…”

Additional Topics

“…Vector quantisation has been widely applied as a preprocessor in many studies: r Speech recognition. Zhang et al (1994) assess three different vector quantisers (including the LBG algorithm and an algorithm based on normal mixture modelling) as preprocessors for a hidden Markov model based recogniser in a small speech recognition problem. They found that the normal mixture model gave the best performance of the subsequent classifier.…”