摘要
为了满足数字通信及其他商业应用的需求,语音压缩编码技术得到迅速发展。近年来主流的低速率语音编码方案主要基于LPC-10,混合激励线性预测(MELP),多带激励编码(MBE),正弦变换编码(SCI),波形内插编码(WI),大多都工作在2.4 kb/s速率下。作为一种重要的低速率语音编码算法,MELP算法对LPC-10编码方案进行大量改进,引入混合激励,非周期脉冲,残差付氏幅度谱,脉冲散布和自适应谱滤波5个特征。实验结果表明,该混合激励线性预测编码在2.4 kb/s上得到了更好的合成语音,并使得合成语音能更好地拟合自然语音。
In order to satisfy demands of the digital communication and other commercial applications, the speech compression technology has been developed rapidly.There are several main LBRSC prevailing in recent years,such as line prediction coding （LPC- 10）, mixed excitation line prediction （MELP）, multi-band excitation （MBE）,sinusoidal transform cording （STC） ,and waveform interpolation（WI）,most of which works at the rate of 2.4kb/s.MELP algorithm is based on the basis of Line Prediction （LP）.Five characters has been introduced into MELP algorithm,they are mixed excitation,a periodic pulse, Fourier magnitudes pulse dispersion and adaptive spectral filtering.The experimental results show that the mixed ex citation linear predictive coding in the 2.4 kb/s gives a better synthesized speech,and makes the synthesized speech to better fit the natural voice.
出处
《电子设计工程》
2010年第3期25-28,共4页
Electronic Design Engineering
