Robust speaker verification in low bit rate channels

The influence of speech coding on text-independent speaker verification was studied.Two methods were proposed to improve the robustness of systems under the matched condition and mismatched one,respectively.The weighted LPCC feature based on the distortion of speech coding was explored to reduce the influence under the matched condition.Under the mismatched condition,the solution was to construct multi-models in multi-channels.In this method,a GMM speech coder detector was used to determine the pattern of the testing speech,and then the corresponding speaker verification model was chosen.Experiments show that the two methods can improve the performance of speaker verification in low bit channels.