Robust speech recognition with selective input data to a NN classifier

This paper presents the new robust IWSR systems which provide selective input data to a neural network in response to speech signal to acoustic noise ratios to improve speech recognition system performance The proposed robust isolated word speech recognition systems employ FMQ/MQ as the spectral labelling process, followed by a Hidden Markov Model (HMM), or a HMM and Neural Network (HMM/MLP) classification techniqut. NOSEX92 has been used as the speech and noise database. This robust performance ensures that a recognition accuracy of the order 98.9% and 95% is obtained with SNR at 10 dB and 0 dB respectively.