摘要
提出了一种面向窄带通信的极低速率参数语音编码算法。在2.4kbps MELP标准的基础上结合听觉感知,对线谱对参数进行联合矢量量化、对基音周期进行内插和非线性量化、对能量参数进行高效压缩,可以使语音数据在0.5kbps下匀速传输;线谱对参数的预测残差用于矢量量化,这是一种提高合成语音的音质的有效方法。实验结果表明,采用本文提出的语音编码算法可以使语音数据在极低码率下有效的传输,解码端合成的语音具有较高的可懂度。
A speech coding algorithm applied to narrowband transmission at very low bit rate is proposed. Based on the MELP standard working at 2.4kbps, joint vector quantization is applied to linear spectrum pair' s parameters, interpolation and nonlinear quantization are applied to pitch, gain parameters are compressed efficiently and auditory perception is con- sidered, speech data could be coded and transmitted smoothly at 0.5kbps The predicative residuals of linear spectrum pair's parameters are also used for vector quantization; it' s a suitable way of improving synthetic speech quality. The experimen- tal results reveal that the proposed algorithm can ensure speech effective transmission at very low bit rate while the intelligi- bility of received speech is acceptable.
出处
《信号处理》
CSCD
北大核心
2013年第9期1134-1141,共8页
Journal of Signal Processing
基金
国家自然科学基金(61273288,61233009,61203258,61011140075,90820303)
中国新加坡数字媒体研究院资助项目(CSIDM)
关键词
联合矢量量化
非线性量化
预测残差
听觉感知
joint vector quantization
nonlinear quantization
predicative residuals
auditory perception