SU591908A1 - Speech signal segmenting device - Google Patents

Description

(54) SPEAKER SIGNALS SEGMENTATOR

one

The invention relates to the field of acoustics, in particular, to devices for automatic recognition of speech patterns, and can be used in cybernetics and communication.

Devices are known that use speech patterns to identify specific acoustic features that characterize individual phonemes or phoneme groups 1.

Their disadvantage is the complexity of the nodes and schemes included in them, due to the specific features of speech used.

Devices are also known in which the selection of a particular class of phonemes, in particular, tonal, is made by phase-shifting the frequency components of a speech signal, by separately detecting the positive and negative half-waves of the speech envelope and comparing the extracted signals in amplitude 2.

The predominance of one of the half-waves (asymmetry) of the envelope serves as a parameter characterizing the presence of a tonal (voice) phoneme.

These devices contain a speech source, a series-connected band-pass filter, a phase-shifting circuit, positive and negative half-wave detectors connected in parallel, an algebraic adder and a switching device, to the outputs of which the control inputs of the key circuits are connected, and the second inputs of the latter are connected to the output of the speech source signal.

The disadvantage of this device is that it allows you to select only phonemes of one class, which reduces the information content of the parameters of the processed speech signals.

The purpose of the invention is to increase the informativity of the selected speech signal parameter.

This is achieved by introducing a correction circuit into the known device, the input of which is connected to the output of the speech source and the output to the input of the band-pass filter.

The drawing shows a block diagram of a speech signal segmenter. It includes a speech signal source I, a correction circuit 2, a band-pass filter 3, a phase-shifter circuit 4, a detector 5 of positive half-waves, a detector 6 of negative half-waves, an algebraic adder 7, a switch 8 and key circuits 9 and 10.

The device works as follows.

In the initial state, key schemes 9 and 10 are closed. Coming to them natural

A speech signal (voltage or current proportional to the sound pressure wave) cannot reach the outputs of the device. The same speech signal is received by the correction circuit 2, with which it is boosted at high frequencies. Then, the informative frequency region of the signal is cut out by a band-pass filter 3 and fed through phase-shifting circuit 4 to the detectors of positive 5 and negative 6 half-waves. Smoothed different polarity signals are compared in amplitude with an algebraic adder, the output of which forms a bipolar function of the asymmetry of the envelope of the speech signal. with a negative signal - the key circuit 10, the output of which passes the signal, classified as tonal sound. Thus, with the introduction of an additional conversion of the dynamic range of the input signal, the capabilities of the device using the asymmetry of its envelope as a specific speech characteristic are enhanced in the sense of increasing the information content of the selected parameter and improving the speed of the circuit that retains the previous noise immunity.

Claims (2)

1. USSR author's certificate number 20616 Cl. G 10 L 1/02, 1967. 2. US patent No. 3377428, cl. 179-1, 1968.