应用DSI相关量化的多描述语音编码方法

Multiple Description Speech Coding Realized by Distributed Subframe Interleaving with Correlated Quantization

提出一个新的多描述编码方案,它利用分布式子帧交织,生成两个基于码激励线性预测(CELP)的语音描述,从而在解码端利用交织的相关性来实时还原差错信号。其中的相关编码原理是:首先把两个相邻语音分组的线谱对(LSP)参数交织到不同的描述中,并结合参数内插方法来保证重要数据的恢复;其余的子帧参数根据分布式原则分配至相应的描述信号,最终通过差错隐藏技术来重建丢失信号。实验结果证明,本文方法在随机信道下提供了更好的语音质量,在突发错误信道下保证了可接受的实时通话质量,更适合不可靠分组网络的语音传输。

A novel multiple description scheme is presented.The scheme uses a distributed subframe interleaving mechanism to create two code excited linear prediction（CELPs） based multiple descriptions（MD）.Then,error bits are recovered by using correlations between interleaving streams in the decoder.The fundamental correlated interleaving is introduced as follows：（1） linear spectrum pair（LSP） parameters combined with interpolation are interleaved into different descriptions to preserve the most important data.（2） Other parameters belonging to different subframes are distributively assigned for easy error concealment.Experiments demonstrate that the proposed codec has better performances.It ensures an acceptable real-time voice quality in burst channels and is more suitable to unreliable packet networks.