A Robust Speech Recognition Based on the Feature of Weighting Combination ZCPA

This paper presents a new approach to extract anti-noisy speech feature: weighting combination zero-crossings with peak amplitudes, which is based on auditory model. It is an improved model of zero-crossings with peak amplitudes. This approach uses the speech signal and its difference signal as input. The frequency information of speech signal is obtained by upward-going zero-crossing intervals, and the intension information is incorporated by compressing nonlinearly amplitudes. The speech feature is weighted according to the auditory characteristics by using weighting function, and then the output feature is obtained. The recognition part uses HMM. Experimental results demonstrate that this new feature is more robust than the old feature in noise environment