An experimental study of noise on the performance of a low bit rate parametric speech coder

The influence of noise on the performance of a low bit rate parametric speech coder is addressed in this paper. MELP vocoder is used to estimate three parameters: the fundamental frequency, voicing and linear prediction coefficients. Influence of different noises under various acoustic environments on the MELP vocoder's parameters is studied. Pitch accuracy rate, voicing decision error rate and average spectral distortion are used to quantitatively evaluate the corrupted speech quality and intelligibility. The results show that the pitch accuracy and voicing decision are susceptible to the noise even at high SNR and illustrate that the presence of noise in unvoiced frames has great impact on the accuracy of voicing decision. The research will be helpful in designing specific speech enhancement algorithm for low bit rate coding.