基于多带的2.4 kbit/s波形内插算法

由于传统特征波形内插语音编码算法对特征波形相位信息的忽略，以及对特征波形的整体对齐，往往造成语音高频谐波分量丢失，从而导致语音的噪声感。为了提高合成语音的质量，该文引入语音多带清浊音标志，并以此为依据对波形内插编码模型中的慢渐变波形和快渐变波形的相位谱进行估计，在语音合成时则对特征波形采取部分对齐的方法，最后提出了一种基于多带的2.4 kbit/s特征波形内插算法。与传统算法相比，新算法明显提高了语音的清晰度。与标准2.4 kbit/sMELP算法相比，该算法合成语音质量亦略显优势。

Multiband Waveform Interpolation Speech Coding Algorithm at 2.4 kbit/s

In the traditional characteristic waveform interpolation speech coding, the high frequency harmonics of the synthetic speech are usually lost, which makes the speech feel noisy, due to the phase information neglect of the characteristic waveform and the whole characteristic waveform alignment. In order to improve the synthetic speech quality, the multiband surd/sonant flag is first introduced. The phase spectrum of the slowly/rapidly evolving waveform is estimated with the waveform interpolation speech coding model depending on the multiband surd /sonant flag. Then the partial characteristic waveform alignment is used in the speech synthesis section and a 2.4 kbit/s multiband waveform interpolation speech coding algorithm is proposed finally. Compared with the traditional characteristic waveform algorithm, the new algorithm can distinctly improve the speech definition. Compared with the standard 2.4 kbit/s MELP algorithm, the synthetic speech quality is also slightly better.