Method for measuring voice source parameters for linear predictive speech coding systems

In the context of the current direction of research in the fi eld of acoustic measurements – non-invasive analysis of the voice source – the problem of measuring excitation parameters for a vocoder with linear prediction is considered. The acute problem of high computational complexity of known methods of its solution based on the technique of “analysis by synthesis” is indicated. In order to overcome this problem, a high-speed acoustic measurement method has been developed based on the criterion of the minimum average sample value of the linear prediction error. It is shown that this criterion implements the principle of minimizing the energy consumption of the announcer for the speech production. An example of technical implementing the developed method is considered, and estimates of its computational complexity are given. It is shown that, compared to the well-known method of multi-pulse excitation of a linear prediction vocoder using two address books: adaptive and stochastic, the costs of implementation of the proposed method are reduced by several orders of magnitude. To confi rm this conclusion, a natural experiment was conducted using the author's software on a set of vowel phonemes from a control speaker. It is shown that by optimizing the excitation signal shape, the mean sample value of the linear prediction error is signifi cantly reduced. The obtained results can be useful in developing new and upgrading existing systems and technologies for speech coding and synthesis, mobile speech communication and other applications of digital speech signal processing with data compression based on the linear prediction model.

speech signal, speech acoustics, speech tract, vocal source of speech, glottis, speech synthesis, speech coding