CELP speech coding based on mel-generalized cepstral analyses

✍ Scribed by Kazuhito Koishida; Keiichi Tokuda; Takao Kobayashi; Satoshi Imai

• Book ID: 101294860
• Publisher: John Wiley and Sons
• Year: 2000
• Tongue: English
• Weight: 215 KB
• Volume: 83
• Category: Article
• ISSN: 1042-0967
• DOI: 10.1002/(sici)1520-6440(200005)83:5<32::aid-ecjc4>3.0.co;2-a

✦ Synopsis

In this paper, a CELP speech coding scheme based on mel-generalized cepstral analysis is proposed. In this scheme, speech spectra can be represented by mel-generalized cepstra, the representation accuracy of the zero and poles of the spectra can be varied continuously from allpole model to cepstral model, and perceptual characteristics of humans regarding the frequency resolution can be considered. The encoding and decoding processes of the proposed method are expected to be suitable for human perception since perceptual weighting and postfiltering operations can be performed through mel-generalized cepstral coefficients. The proposed scheme and the conventional CELP scheme based on linear prediction were compared subjectively for a bit rate of 4 kbit/s. The results showed that the proposed scheme outperforms the conventional method by about 2 dB for an equivalent Q value.