论文部分内容阅读
为解决正弦激励线性预测(SELP)2.4 kb/s语音编码算法中清浊音过渡时合成语音的质量较差的问题,该文利用子带清浊音判决参数量化过程中存在的冗余度,提出了一种跳跃帧判决及处理方法。该方法根据当前帧前后各60个样点的平均能量比值判定当前帧是否为跳跃帧,并利用子带清浊音判决参数量化的冗余度传送跳跃帧信息。解码端根据当前帧是否为跳跃帧对解码端参数采用不同的插值方案。测试结果表明,该方法能够将合成语音的M O S分提高0.004左右。
In order to solve the problem of poor quality of synthesized speech in the process of voiced speech transition in the 2.4 kb / s speech coding algorithm with sinusoidal excitation linear prediction (SELP), this paper uses the sub-band unvoiced and voiced speech to judge the redundancy existing in the quantization of parameters, A Jump Frame Decision and Processing Method. The method determines whether the current frame is a skip frame according to the average energy ratio of 60 samples before and after the current frame and transmits the hopping frame information by using the redundancy quantified by the sub-band voicing decision parameter. The decoder uses different interpolation schemes for the decoder parameters according to whether the current frame is a skip frame or not. The test results show that this method can improve the MOOS of synthetic speech by about 0.004.