论文部分内容阅读
为了更好地区分正弦激励线性预测声码器中待量化基音周期参数矢量各维基音周期参数的重要性,提高合成语音质量,提出了一种新的权重系数计算方法。发现权重系数的计算应考虑当前语音帧所有子带的清浊音度参数,而不仅仅是第一个子带的清浊音度参数,这样能够更好地区分各语音帧基音周期参数的重要性。将各语音帧的所有子带清浊音度参数相加,其和作为基音周期参数量化时的权重系数。测试结果表明,与仅将第一个子带的清浊音度参数作为权重系数相比,新方案能使合成语音质量MOS有较大提高。
In order to better distinguish the importance of each pitch period parameter of the quantized pitch period parameter vector in sinusoidal excitation linear prediction vocoder and improve the synthesized speech quality, a new method of calculating the weight coefficient is proposed. It is found that the weight coefficient calculation should consider the voicing degree parameters of all the sub-bands of the current speech frame instead of the voicing degree parameters of the first sub-band, so as to better distinguish the importance of the pitch period parameters of each voice frame. The sum of the voiced and unvoiced parameters of all the sub-bands of each voice frame is added, and the sum is used as the weight coefficient when the pitch period parameter is quantized. The test results show that the new scheme can greatly improve the quality of synthesized speech (MOS) compared with using only the first sub-band’s voicing parameters as the weight coefficients.