SU781887A1

SU781887A1 - Device for segmentation of speech signal

Info

Publication number: SU781887A1
Application number: SU792738565A
Authority: SU
Inventors: Николай Степанович Куцоконь
Original assignee: Ульяновский политехнический институт
Priority date: 1979-03-13
Filing date: 1979-03-13
Publication date: 1980-11-23

Description

Изобретение относится к приборостроению и может быть использовано в акустике, в частности·, при автоматическом распознавании образов в кибернетике и связи. , 5The invention relates to instrumentation and can be used in acoustics, in particular, · with automatic pattern recognition in cybernetics and communications. , 5

Известен сегмент’атор речевых сигналов, содержащий источник речевого сигнала, последовательно соединенные полосовой фильтр, фазосдвигающую схему, включенные параллельно друг Ю другу детекторы положительных и отрицательных полуволн, алгебраический сумматор и коммутирующее устройство, к выходам которого подключены управляющие входы ключевых схем, вторые 15 входы последних соединены с выходом источника речевого сигнала, а также корректирующую схему, вход которой подключен к выходу источника речевого сигнала, а выход - к входу полосово- 20 го фильтра £1] .A known segmentator of speech signals containing a source of speech signal, a series-connected bandpass filter, phase shifting circuit, positive and negative half-wave detectors connected in parallel to each other, an algebraic adder and a switching device, the outputs of which are connected to the control inputs of key circuits, the second 15 inputs of the last connected to the output of the source of the speech signal, as well as a correction circuit, the input of which is connected to the output of the source of the speech signal, and the output to the input of the strip 20th filter £ 1].

Однако помехоустойчивость этого устройства недостаточно высокая, а наличие импульсной помехи часто учитывается устройством как асимметрия 25 огибающей, что снижает достоверность распознавания фонем.However, the noise immunity of this device is not high enough, and the presence of impulse noise is often taken into account by the device as an asymmetry of the envelope 25, which reduces the reliability of recognition of phonemes.

Наиболее близким к предлагаемому является устройство сегментации речевого сигнала, состоящее из анализаΰ 2 · тора, выполненного на выходе с час-. тотными каналами, каждый из которых содержит ключевую и пороговую схемы [2] .Closest to the proposed device is a segmentation of a speech signal, consisting of an analyzer ΰ 2 · torus, performed at the output from the hour. total channels, each of which contains a key and threshold circuit [2].

Недостаток этого устройства в недостаточно высокой информативности выделяемых параметров речевого сигнала.The disadvantage of this device is the insufficiently high information content of the allocated parameters of the speech signal.

Цель изобретения - повышение информативности выделяемых параметров речевого сигнала.·»The purpose of the invention is to increase the information content of the allocated parameters of the speech signal. · "

Поставленная цель достигается тем, что каждый канал снабжен последовательно соединенными детектором, интегрирующим амплитудным 'модулятором и блоком ср’авнения, включенными между выходом анализатора и первым входом ключе.вой схемы, блоком задержки, включенным между этим же выходом анализатора и вторым входом ключевой схемы, а также импульсным генератором, соединенным с выходом пороговой схемы и вторым входом модулятора, при . этом вход пороговой схемы соединен с входом детектора и с указанным выходом анализатора.The goal is achieved in that each channel is equipped with a series-connected detector, an integrating amplitude 'modulator and a comparison unit connected between the output of the analyzer and the first input of the key circuit, a delay unit connected between the same output of the analyzer and the second input of the key circuit, as well as a pulse generator connected to the output of the threshold circuit and the second input of the modulator, at. the input of the threshold circuit is connected to the input of the detector and to the specified output of the analyzer.

На чертеже изображена структурная схема предлагаемого устройства.The drawing shows a structural diagram of the proposed device.

Выходной сигнал каждого частотного канала анализатора 1 подается на вход детектора 2, детектируется и поступает на вход интегрирующего амплитудного модулятора 3, со второго входа которого производится сброс проинтегрированного сигнала импульса’ ми генератора 4 стабильной вольтсе“’•““’ЙУЙЙ'ЙОЙ' площади $₀ , причем ъThe output signal of each frequency channel of the analyzer 1 is fed to the input of detector 2, detected and fed to the input of the integrating amplitude modulator 3, from the second input of which the integrated signal is pulsed by the pulses of the stable voltage generator 4 “” • ““ ' ₀ , and b

max >.max>.

где U (t) - входной сигнал. -.-/-===.--===Проинтегрированный за время ΐ' й промодулированный сигнал подается на вход блока 5 сравнения, в котором сигналы разделяются на прямой и 'задержанный на время t и сравниваются по амплитуде. При этом в каждый момент времени проверяется неравенствоwhere U (t) is the input signal. -.- / - === .-- === The time-integrated modulated signal is fed to the input of the comparison unit 5, in which the signals are divided into direct and delayed by time t and compared in amplitude. Moreover, at each moment of time, the inequality

Чтобы исключить необходимость варьирования частоты, опроса при изменений интервала корреляции сигнала, интервал интегрирования выбирается в соответствии с минимальным интервал0МК©&Е5ёляций7 Благодаря этому возможность выдачи ложного сегмента исключается.To eliminate the need for frequency variation, polling for changes in the signal correlation interval, the integration interval is selected in accordance with the minimum interval 0MK © & Е5ёlyatsiy7 Due to this, the possibility of issuing a false segment is excluded.

Использование предлагаемого устройства позволяет значительно повысить Точность выделения сегментов перехода и исключить возможность пропуска сегментов и выделения лишних сегментов перехода, что значительно повышает информативность выделяемых параметров речевого сигнала и, следовательно , достоверность распознавания речевых сигналов.Using the proposed device can significantly increase the accuracy of the selection of transition segments and eliminate the possibility of skipping segments and highlighting unnecessary transition segments, which significantly increases the information content of the selected parameters of the speech signal and, therefore, the reliability of recognition of speech signals.

где U_o t;where U _o t;

I У b^dt-y U(t)dtb/U_oi I Y b ^ dt-y U (t) dtb / U _oi

4^ Ч порог срабатывания блока 5 сравнения, зависящий от среднего интервала корреляции речевого Сигнала и уровня шумов канала, время интегрирования.4 ^ H the threshold of the block 5 comparison, depending on the average correlation interval of the speech signal and the noise level of the channel, the integration time.

Claims

(51) SPEECH SEGMENTATION DEVICE The invention relates to instrument making and can be used in acoustics, in particular, in automatic pattern recognition in cybernetics and communication. The known speech signal segmenter contains a speech source, a series-connected band-pass filter, phase shift eldyu circuit, detectors of positive and negative half-waves connected in parallel, an algebraic adder and a switching device to the outputs of which are connected the control inputs of key circuits, the second inputs of the latter are connected to the output speech signal source as well as a correction circuit whose input is connected to the output of the voice signal source and the output to the input of the band tra l. However, the noise immunity of this device is not high enough, and the presence of impulse noise is often taken into account by the device as an envelope asymmetry, which reduces the reliability of recognition of phonemes. The closest to the present invention is a speech signal segmentation device, consisting of a SIGNLL torus analysis performed at the output from the clock. total channels, each of which contains the key and threshold schemes 2. The disadvantage of this device is not enough high information content of the selected speech signal parameters. The purpose of the invention is to increase the awareness of selected speech parameters. , The goal is achieved by the fact that each analog is equipped with a series-connected detector that integrates an amplitude modulator and a comparison unit, connected between EXIT / 1OM, and a delay unit, connected between the same analyzer output and the second input of the key circuit, also a pulse generator connected to the threshold circuit and the second modulator input, at. This input threshold circuit is connected to the input of the detector and with the specified output of the analyzer. The drawing shows a structural diagram of the proposed device. The output signal of each frequency channel of the analyzer 1 is fed to the input of detector 2, detected and stepped at the input of its amplitude modulator 3, from the second input of which the integrated pulse of generator 4 of stable volts K of the square of So r is released and rU (t) c (t where u (t) is the input signal.: g,. ™ .... ...:. The time-integrated and modulated signal is fed to the input of comparator unit 5, in which the signals are separated into direct and -e - held at time t and compared with amplitude. The first moment of time tr is verified as the inequality t ;: -tnt 1 rg IJ UCt) dt-J U (.t) dt where JQ is the response threshold of the comparison unit 5, depending on the average correlation interval of the speech signal and the level of noise and nal aa - integration time. If the inequality is fulfilled, the comparison block 5 generates a pulse of duration C, which is opened by the key eByio cxeiMy 6, through which the delayed input segment of the device 7 is delayed by the element 7 of the device. The delay is necessary to compensate for the integration time. The segmenter is triggered by a threshold circuit 8, the trigger threshold of which corresponds to the channel level. .. Ksdolnyy in the proposed seg Mentator of the integrating amplitude modulator allows not to compare the signal strengths, but their integral sums. At the same time, at each time interval t, cffSTWeTCT during the integration time, there is a continuous on-line formation of a transition segment. Therefore, the probability of a segment loss is zero to zero and a high frequency of polling channels is not required. In order to EXCLUDE the need to vary the frequency of polling when the signal correlation interval changes, the integration interval is selected in accordance with the minimum interval; ke1) Good Gee 7 Thanks to this, the possibility of issuing a false segment is excluded. The use of the proposed device allows you to significantly improve the accuracy of selection of transition segments and eliminate the possibility of skipping the needles and extracting unnecessary transition segments, which significantly increases the efficiency of the selected parameters and, consequently, to improve the recognition of speech signals. The invention The speech signal segmentation device, consisting of an analyzer, performed at the output with frequency channels, of which contains a key and threshold circuit, characterized in that, in order to increase the informativity of the extracted speech signal parameters, each channel is equipped with successively matched a torus, and an amplifying modulator and a comparison unit, which are switched on between the analyzer output and the first input of the key circuit, a delay unit included between the same analyzer output and the second input of the key circuit, as well as the pulse generator (er5m, connected to the output of the threshold of the circuit and the second input of the module, while the input of the threshold circuit is connected to the input of the detector and with the output of the analyzer. Considering the information taken into account during the examination of 1 Lvtr the witness of the USSR btw 591908, class G 10 L 1 / (2, 1976.

2. USSR author's certificate 585520, cl. G 10 L 1/02, 1976 (protype).

.i-: S :, - -