An efficient coding using the correlation between time series of differential LSP parameters
Shuichi Arai, Shinji Ozawa · Electronics and Communications in Japan (Part III Fundamental Electronic Science) · 1989
Abstract The information compression by LSP analysis‐synthesis is a promising method in the sense that speech with relatively high quality can be synthesized with a small number of codes. Up to now, coding of LSP parameters has been investigated from various viewpoints. LSP analysis is performed framewise independently, while rarely does the speech information change rapidly with time. Consequently, when viewed as a time‐series, LSP parameters contain redundancies. On the other hand, LSP parameters are ordered in the ascending order of the frequency, and the time‐series exists for each parameter. A correlation exists between two time‐series which are adjacent on the frequency axis. From those viewpoints, this paper examines first the statistical features of LSP parameters when viewed as the time‐series. the coding based on those features is proposed, and the essential usefulness of the method is verified experimentally. Then, a new highly efficient coding method is proposed by combining the two foregoing ideas. In the proposed method, the interframe difference of LSP parameters is encoded in the first stage, which can reduce the coding complexity. In the second stage, the coding complexity required for transmission is reduced further by encoding the difference of the time‐series adjacent on the frequency axis. Applying the proposed system, coding experiment and the quality evaluation of synthesized speech were performed, and the validity of the proposed method was confirmed.