Method of compression of voice data without a pause based on the optimal level quantization coefficients of the expansion pieces of speech signals for eigenvectors subband matrix

Existing techniques of compression of speech sounds without pauses, using a coarse quantization level based on a psychoacoustic model, which results in the need for the so-called subband transforms segments (vectors) samples of speech signals, allowing to obtain the other vectors subvectors which reflect the frequency properties of the input vector in the selected ranges frequency axis. That component of the sub-vectors is quantized by the level of different steps, thus achieving the accounting frequency-selective properties of human hearing. Currently, for a subband common to use transformation process output sequences decimation FIR filters (finite impulse response) tuned to respective portions frequency axis. This procedure subband transform is not optimal in the sense of minimizing the approximation error spectra initial vectors to selected frequency ranges, which leads to increased errors in data recovery on the quantized values and, as a consequence, degradation of the reproduced speech. In this connection, the authors proposed the speech data compression method without pauses, created with the use of mathematical apparatus developed on the basis of eigenvectors subband matrix, allowing adequately formulate variational conditions and solve optimization of voice data processing tasks. The proposed method of speech compression without pauses allows data depending on the starting bit of said data samples provide compression ratios up to 12 times.