Complexity reduction and quality improvement of code-excited linear predictive coders부호여기 선형예측부호화기의 연산량 감축 및 음질 향상에 관한 연구

Cited 0 time in webofscience Cited 0 time in scopus
  • Hit : 340
  • Download : 0
The main objective of this dissertation work is to develop a CELP coder that is computationally more efficient and has improved output speech quality as compared to the existing coders. To reduce the computational complexity of the CELP coder, two attempts are made. One is to use a fast codebook search method, and the other is to investigate a new low-rate SELP coder that is based on the structure of the CELP coder but does not require codebook search. As a scheme to improve the output speech quality of the CELP coder, we propose a method that performs residual shaping both at the analyzer and at the synthesizer. In CELP coding, the analysis-by-synthesis method is used to obtain highquality output speech at low-bit rates. When this method is used in connection with a large codebook, a large number of computations are required in the encoding phase. To solve this problem, we develop a computationally efficient method for codebook search in CELP coding. By combining the sparse code vector and the frequency-domain search method simultaneously, we can reduce the computational complexity to an extent that cannot be obtained by using either method separately. It has been shown that the speed-up factor of the proposed method is almost twice that of the frequency-domain search method that is so far known to be the fastest method. Since our proposed codebook search method is suboptimal, the obtained SNR value may be somewhat lower than those of the full search method. However, according to our informal listening tests, the difference in speech quality between the original search method and our proposed method is almost indistinguishable. Next, as another scheme to reduce the computational complexity of the CELP coder, we present a multi-stage SELP coder that can be used as a speech coder with low-bit rate. The multi-stage SELP coder has the form of multistage vector quantization, but utilizes the structure of a conventional SELP coder. The proposed multi-stage SELP code...
Advisors
Un, Chong-Kwanresearcher은종관researcher
Description
한국과학기술원 : 전기 및 전자공학과,
Publisher
한국과학기술원
Issue Date
1990
Identifier
61503/325007 / 000815215
Language
eng
Description

학위논문(박사) - 한국과학기술원 : 전기 및 전자공학과, 1990.2, [ ix, 118 p. ]

URI
http://hdl.handle.net/10203/36105
Link
http://library.kaist.ac.kr/search/detail/view.do?bibCtrlNo=61503&flag=dissertation
Appears in Collection
EE-Theses_Ph.D.(박사논문)
Files in This Item
There are no files associated with this item.

qr_code

  • mendeley

    citeulike


rss_1.0 rss_2.0 atom_1.0