Метод кодирования голосового источника турбулентного типа на основе гибридной модели линейного предсказания

Измерительная техника. 2026. Т. 75. № 3. С. 105–113.

Савченко В. В., Savchenko L.

Within the framework of a current area of research in the field of speech acoustics – non-invasive analysis of speech
production processes – the acute problem of insufficient accuracy of parametric methods for coding a turbulent (noise) type voice source is considered. In order to overcome this problem, a method for coding a sound source with increased accuracy has been developed, based on a hybrid model of linear speech prediction, which combines the advantages of parametric and nonparametric approaches to speech signal modeling. In this case, the parametric approach is implemented in the form of a vector of linear prediction coefficients, and the nonparametric approach is implemented in the form of a clipped sequence of linear prediction error samples. Using the author's software, a
full-scale experiment on a set of the whispered speech sounds of the control speaker has been set up and carried out. Compared to a known method for encoding a turbulent sound source based on a noise-excited linear prediction model, the developed method is characterized by an accuracy gain of 2.5 dB or more in the mean square error of linear prediction metric, while guaranteeing speaker voice recognition from the decoded (reconstructed) speech signal. The obtained results will be useful in developing low-cost systems and technologies for digital processing, synthesis, and transmission of speech with multiple data compression. Promising applications of the developed method include digital voice biometrics systems, in which speaker voice recognition is a key requirement for the speech signal encoding method 

Keywords: устная речь, шепотная речь, oral speech, speech signal, speech acoustics, vocal tract, Glottis, речевой сигнал, Speech coding, whispered speech, речевая акустика, речевой тракт, голосовая щель, кодирование речи