A speech error concealment algorithm based on parameter redundancy and voiced/unvoiced classification
Xiao Yong, Xiongwei Zhang, Pla Uni · Signal Processing · 2009
To avoid the severe degradation of low-bit rate coded speech when transmitted over noisy communication channels,an error concealment algorithm for MELP voeoder is proposed.One-step Markov chain is used to exploit the interframe and intraframe redundancy. The transfer probability between bandpass voicing parameter and the other parameters is presented.During speech synthesis, an algorithm based on maximum likelihood criterion is proposed to determine the voiced or unvoiced mode.Speech parameters are reconstructed according to minimum square error criterion during voiced frames,while reconstructed directly by mapping during unvoiced frames.Finally,our algorithm is tested under the scale of 10~(-2)bit error rate.Compared with the two recently proposed error concealment algorithms,the performance of the new algorithm is more excellent.