Scholarlink

<p>In the context of non-invasive voice source analysis, a&#xa0;key area of acoustic measurement research, this paper addresses the measurement of excitation signal parameters and waveforms for linear predictive vocoders. The study highlights the critical challenge of high computational complexity in existing analysis-by-synthesis algorithms. In order to overcome this challenge, a&#xa0;high-speed acoustic measurement method has been proposed based on the criterion of minimum mean-square linear prediction error. This criterion is shown to reflect the principle of minimizing the energy consumption of a&#xa0;speaker during speech production. An example of technical implementation of the method is presented alongside an evaluation of its computational complexity. It is shown that compared to a&#xa0;standard method of multi-pulse excitation of a&#xa0;linear predictive vocoder using two address codebooks (adaptive and stochastic), the implementation costs of the proposed method are reduced by dozens of times. To validate these findings, a&#xa0;full-scale experiment was conducted using proprietary software on a&#xa0;set of vowel phonemes from a&#xa0;control speaker. It is shown that optimizing the excitation signal waveform significantly reduces the mean-square linear prediction error. Our results can be useful when developing new and upgrading existing systems and technologies for speech coding and synthesis, mobile speech communication, and other digital speech signal processing applications with data compression based on the linear prediction model.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Measuring voice source parameters in linear predictive speech coding systems

  • 期刊论文

  • 发表日期：

    2026年5月6日

查看全文
  • V. V. Savchenko,
  • L. V. Savchenko

摘要

In the context of non-invasive voice source analysis, a key area of acoustic measurement research, this paper addresses the measurement of excitation signal parameters and waveforms for linear predictive vocoders. The study highlights the critical challenge of high computational complexity in existing analysis-by-synthesis algorithms. In order to overcome this challenge, a high-speed acoustic measurement method has been proposed based on the criterion of minimum mean-square linear prediction error. This criterion is shown to reflect the principle of minimizing the energy consumption of a speaker during speech production. An example of technical implementation of the method is presented alongside an evaluation of its computational complexity. It is shown that compared to a standard method of multi-pulse excitation of a linear predictive vocoder using two address codebooks (adaptive and stochastic), the implementation costs of the proposed method are reduced by dozens of times. To validate these findings, a full-scale experiment was conducted using proprietary software on a set of vowel phonemes from a control speaker. It is shown that optimizing the excitation signal waveform significantly reduces the mean-square linear prediction error. Our results can be useful when developing new and upgrading existing systems and technologies for speech coding and synthesis, mobile speech communication, and other digital speech signal processing applications with data compression based on the linear prediction model.