A detection method of unbearable bit error rate for a 2.4kbps MELP vocoder

In harsh channel conditions, the quality of the synthetic speech at low bit rate would be affected severely. In order to improve the robustness of the vocoder and make it more resilient to errors in random channel, unequal error protection (UEP) channel coding is usually adopted. However, when the errors cannot be corrected in some cases, UEP channel coding will not improve the quality of the synthetic speech. More seriously, the quality of the synthetic speech may deteriorate rapidly. In order to improve the user's feeling in such case, a detection method of unbearable bit error rate (BER) is proposed in the paper. This method can increase the error resilience capability of the vocoder in random channel. When the error bits cannot be corrected, the vocoder will output comfortable noise to improve the user's feeling. This method is used for a 2.4 kbps vocoder based on MELP, and the test results show that it works effectively.