SU781882A2 - Word identification device - Google Patents

Description

(-vU DEVICE FOR LAYER RECOGNITION

The invention relates to instrumentation and is used for speech control in human-machine systems. According to the main author. St. No. 516094 a device for recognizing words is known, comprising a microphone connected to an envelope correction unit through a speech clipper's clipped envelope selector, and a larin keyboard that is connected through a highlight to the second input of the envelope correction unit, the outputs of which are connected to the selector inputs of ring sounds deaf sounds, ringing bows and deaf bows; and with the selection word selector, the outputs / selections of the tori are connected to the inputs of the codec unit, and the outputs are encoder The signs and the output of the word envelope are connected to the inputs of the decision block. However, in the practical application of the device under the conditions of intense acoustic interference, the codes of many widely used words appear close to the interference codes, which causes false positives of the device. The purpose of the invention is to improve the device, ensuring a reduction in the number of false positives. The goal is achieved by introducing a word length selector into the device, the input of which is connected to the auxiliary output of the word envelope selector, and the outputs are connected to the additional inputs of the decision block. The drawing shows a diagram of the device. The device contains a microphone 1 and a laryngophone 2, which through a selector 3 of the clipped envelope1; to it a speech signal and the selector 4 of the clipped envelope of the main tone are connected to the inputs of the correction-envelope unit 5. The outputs of block 5 of the correction envelopes, respectively, are connected to the inputs of the selector of voiced sounds, the selector of 7 deaf sounds, the selector of 8 voiced bows of the selector 9 of deaf bows and the selector of 10 word envelopes. The outputs of the selectors are connected to the inputs of the coding block 11, the outputs of which, as well as one main output of the word envelope selector, are connected to the inputs of the decision block 12, in addition ;; The yyyylyyy output of the envelope of the word envelope is connected to the selector 13 of the word length, the outputs of which are connected to the additional inputs of the unit for deciding ...-; -, -; - ri --- 2nd r - kg -f :.

. The device works as follows.

When any words are pronounced, the speech signal perceived by microphone 1 is fed to the selector 3 of the clipping signal. It must be equal to zero before the pronunciation of the phonemes P, T, K, C, H and the phonemes B, D, G. when pronouncing in the word deaf, 33ROPHY and fricative consonant phonemes С, Ш, Ф, X, П, Т, К, Ч, Ц.

 -Y W; :::;:; : :: g: :: -: - - - -

Signals with a selection of signals are received by block 5 of the corr .; of the envelope responses, which forms a corrective response CMYINYYGSHEeRibe speech signal and the main ;;;;;. tones .:.

Corrected clipped envelopes arrivals 10yMyo ya ° 6y LM-tori b - 9 and a 10 times envelope; the words.

Selectors 6-9 form on. clipped signal envelopes

testifying to the presence in pro. iene ennrm of the word corresponding ogg. - -

The selector 10 of the word envelope forms the envelope of the word, the crtor is equal to a constant value during the passage of the word, and its beginning coincides with the appearance of either a curb clipping or clipping envelope of the speech. Sign al a, li boccd sh-dv anvna kl ip pyennoy envelope rsnrvnogo tone; gvzYyYY66sty on how many of them. on is the first.

The symbol encoding unit 11 fixes the aremore sequence of features as they appear. In the layer. .: -. - - -. ; Simultaneously with the beginning of the formation of the envelope, word block 10 gives out I syyal to selector 13 for real word I, in which two thresholds are set; l. &Amp;

one :,

j,:., .., - -

output time intervals: lower and upper. | The outcome of the most short-range words of the worker is the one selected; the upper one is of the first one of the most important. The selector op 13 entails: on a solid output circuit, a logical value of the zero level if the acoustic signal is less for the lower threshold threshold and provides the logical value of the unit level if the acoustic signal is longer than the lower threshold interval. In addition, the selector 13 is at a different output logical value, zero level, if the duration of the acoustic signal is less than the duration of the upper threshold interval, and the logical value of the unit level - in the right case.

Distinction of the spoken word is made with the help of decision making unit 12, representing the codeVA comparison scheme recorded in feature coding unit 11, and the set of reference letters. The decision about the spoken word is made only at the end of envelope 1; duration selector cJiova ..,. .

Using the word length selector allows you to include words that are similar in noise codes but differ in duration in the working vocabulary, t, a. a simple way to increase the number of recognizable words; to increase the reliability of its work by reducing the number of false trials from external acoustic interference.

Claims (1)

1. Copyright Soviet Union №516094,. cl. G ЮТ i / ОО; 19:74;