A New Method Towards Speech Files Local Features Investigation

There are a few reasons for the recent increased interest in studying the local features of speech files. It is stated that many of the essential properties of the speaker’s language used may appear in the form of a speech signal. The traditional instruments — short Fourier transform, wavelet transform, Hadamard transforms, autocorrelation, and the like can detect not all peculiar properties of the language. In this paper, we propose a new approach to such characteristics exploration. The original signal is approximated by a new one, which values are taken from a finite set. We then construct a new sequence of fixed-size vectors based on these approximations. Studying the distribution of generated vectors provides a new way of describing the local attributes of speech files. Finally, the developed technique is applied to the problem of the automated distinguishing of two known languages used in speech files. For this, a simple neural net is constructed.