Robust speech recognition using perceptual wavelet denoising and mel-frequency product spectrum cepstral coefficient features