Very Low Bit Speech Coding Using Fractal Systems.

使用分形系统的极低比特语音编码。

• 批准号: 08650418
• 负责人: KOHATA Minoru
• 金额: $ 1.09万
• 依托单位: Tohoku UNIVERSITY
• 依托单位国家: 日本
• 项目类别: Grant-in-Aid for Scientific Research (C)
• 财政年份: 1996
• 资助国家: 日本
• 起止时间: 1996 至 1997
• 项目状态: 已结题
• 来源: https://kaken.nii.ac.jp/grant/KAKENHI-PROJECT-08650418/
• 关键词: speech coding; Very low bit coding; fractal systems; perceptual coding; auditory filter

A very low bit rate speechcoder at 1.2kbpsis newly proposedln this coder which is similar to the LPC vocoder, speechsignal is synthesizedusing sinusoids whose frequenciesare multiple of the fundamental frequency, and whose amplitudes are adaptively modulated based on auditory perceptualcharacteristics, in order to improve the quality.In this study, the basics of the new sinusoids synthesizerare introduced, thenhow to modulate the sinusoids based on perceptual characteristics is explained. The auditory perceptual characteristics are simulated by Gammatonefilters. And a new segment coder of spectral information (LSP coefficients) using DP matchingis also proposedtoreduce the total bit rate to 1.2kbps. According to MOS test (5 grades), the subjectivequality of the synthesized speech is improved by 1.0 compared with that of the simple LPC vocoder.

本文提出了一种类似于LPC声码器的1.2kbps的极低比特率语音编码器，使用频率为基频倍数的正弦波合成语音信号，并根据听觉感知特征对其幅度进行自适应调制，以提高语音信号的质量。本文首先介绍了新型正弦波合成器的基本原理，然后阐述了如何基于感知特性调制正弦波。用伽玛通滤波器模拟了听觉感知特征。并提出了一种基于DP匹配的频谱信息（LSP系数）分段编码器，将总比特率降低到1.2kbps。经MOS（5级）测试，合成语音的主观质量比简单的LPC声码器提高了1.0。

项目成果

木幡稔: "リカレントニューラルネットワークを用いた補間法とその音声符号化への応用" 電子情報通信学会論文誌. J79-D-II・5. 969-976 (1996)

Minoru Kobata：“使用循环神经网络的插值方法及其在语音编码中的应用”电子信息通信工程师学会学报 J79-D-II·5（1996）。

M.Kohata: "A New Method of Interpolating LSP Cofficients using Recurrent Neural Networks" Procs.of ISSPA96. 1. 314-317 (1996)

M.Kohata：“使用循环神经网络插值 LSP 系数的新方法”Procs.of ISSPA96。

M.Kohata: "A New Method of Interpolating LSP Coefficients Using Recurrent Neural Networks" Procs.of ICSLP 96. 1. 314-317 (1996)

M.Kohata：“使用循环神经网络插值 LSP 系数的新方法”Procs.of ICSLP 96. 1. 314-317 (1996)

M.Kohata: "Interpolation of LSP coefficients using recurrent neural networks" Electronics Letters. 32, 6. 1441-1442 (1996)

M.Kohata：“使用循环神经网络插值 LSP 系数”《电子快报》。

M.Kohata: "A New Method of Interpolating LSP Cofficients using Recurrent Neural Networks" Procs.of ISSPA96. 1. 57-60 (1996)

M.Kohata：“使用循环神经网络插值 LSP 系数的新方法”Procs.of ISSPA96。