Presentation 1996/5/24
Quantization of LSP Parameters using Kalman-Neuro-Training Method
Yoshinori Morita, Hiroyuki Sono, Tetsuo Funada,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) Vector quantization using a codebook has good performance in low bit-rate telephone-band speech codings, however, it has problems that consume many computations and memory storages. In this paper, we propose vector quantization using a neural-network. One of problems of this approach is how to reduce the learning error. We use the Kalman-Neuro. Training(KNT)method for quantization of LSP parameters. In experimental studies, the spectral distortion results in 1.41dB by 24-bit quantization. To evaluate of performance of this method, we apply Karhunen-Loeve transform, and obtain approximately the same distortion.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) speech coding / kalman filter / neural network / LSP / vector quantization
Paper # NC96-4
Date of Issue

Conference Information
Committee NC
Conference Date 1996/5/24(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Neurocomputing (NC)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Quantization of LSP Parameters using Kalman-Neuro-Training Method
Sub Title (in English)
Keyword(1) speech coding
Keyword(2) kalman filter
Keyword(3) neural network
Keyword(4) LSP
Keyword(5) vector quantization
1st Author's Name Yoshinori Morita
1st Author's Affiliation Ishikawa National College of Technology()
2nd Author's Name Hiroyuki Sono
2nd Author's Affiliation Faculty of Engineering, Kanazawa University
3rd Author's Name Tetsuo Funada
3rd Author's Affiliation Faculty of Engineering, Kanazawa University
Date 1996/5/24
Paper # NC96-4
Volume (vol) vol.96
Number (no) 76
Page pp.pp.-
#Pages 8
Date of Issue