去除AMR语音编码器前瞻时延的研究

A Method to Skip Look-Ahead Samples in AMR Speech Codec

为降低自适应多速率(AMR)语音编码器的时延,提出了一种去除其前瞻时延的方法.此方法采用协方差算法计算线性预测系数,然后用线性预测系数外推得到新样点.用新推得的样点取代原来的前瞻样点,从而去除了编码器的前瞻时延.进一步对重建语音质量分别进行了客观评价和主观评价,并且计算了算法复杂度.实验结果表明该方法在去除前瞻时延的同时,重建语音质量接近原始AMR的重建语音质量.

A method was proposed to skip the look-ahead samples in this paper.Firstly,the method used the covariance algorithm to calculate the linear prediction（LP） coefficients.Secondly,LP coefficients were utilized to produce the new samples.Finally,the new samples were used to replace the look-ahead samples.To evaluate the voice quality,the objective evaluation and subjective evaluation were designed respectively and the complexity test was calculated.Experiments show that the reconstructed quality of the proposed method is near to the original AMR codec,and the delay of the proposed method is lower 5 ms than the original method.