A Variable Weighting Factor Based Distortion Measure for Pitch Quantization

Effective quantization of pitch parameter is very important for low bit rate speech coding algorithms. Traditional method uses a constant weighting factor based distortion measure for pitch quantization based on mixed excitation linear prediction (MELP). In this paper, we propose a variable weighting factor based distortion measure for pitch quantization based on MELP. In this method, we combine several consecutive speech frames as a superframe and quantize jointly. We use weighted squared Euclidean distance measure to search the codeword in the codebook, and use the product of five band pass voicing coefficients´ sum and gain parameter to denote the current speech frame´s weighting factor. Compared with the traditional method, simulation results show that our method reduces the pitch distortion of reconstructed speech by almost 3.3% and raises the mean opinion score by almost 0.09(3.5%).