Robust speech recognition using neural networks and hidden Markov models

Abstract

This paper proposes a robust, speaker-independent isolated word speech recognition (IWSR) system (SMQ/HMM-SVQ/HMM)/MLP which combines dual split matrix quantization (SMQ) and split vector quantization (SVQ) pair combined with both the strength of the HMM in modeling stochastic sequences and the non-linear classification capability of MLP neural networks (NN). The system efficiently utilizes processing resources and improves speech recognition performance by using neural networks as the classifier of the system. Computer simulation clearly indicates the superiority over conventional VQ/HMM and MQ/HMM systems with 98% and 95.8% recognition accuracy at 20 dB and 5 dB SNR levels, respectively in a car noise environment, based on the TIDIGIT database.