Butenko Yu.I. – Ph.D. (Tech.), Associate Professor,

Bauman Moscow State Technical University

E-mail: iuliiabutenko2015@yandex.ru

Konopleva A.A. – Student,

Bauman Moscow State Technical University

E-mail: istommi@icloud.com

Automatic recognition using neural networks of written speech is interpreted from the point of view of the variability of fonts and outlines of letters, including different implementations of phonemes, position and characteristics of acoustic receivers, changes in speech parameters. Word boundaries are determined in the process of recognition using a neural network by selecting the optimal sequence of words that best matches the input speech stream according to acoustic, linguistic and pragmatic criteria. It is necessary to identify the recognition thresholds of unreasonable sequences, which are supposed to be considered the most successful number of unreasonable sequences of three sounds - trigrams.

In the article it is considered the possibility of using trigrams to increase the probability for recognition of individual words in the speech stream. It is given the parameters that characterize the systems of automatic speech recognition. Various parameters of speech, which is a challenge in automatic recognition are described. An approach to speech recognition based on the trigrams of the Russian language is proposed. It is given the subjective assessment of the frequency of the trigrams Russian. It is noted that easily spoken trigrams are recognized with higher probability, and the use of trigrams increases the probability of speech recognition quality.

Butenko Yu.I., Konopleva A.A. Methodology for neuron-network technologies in recognition of trigrams. Neurocomputers. 2020. V. 22. № 1. P. 66–76. DOI: 10.18127/j19998554-202001-01.

1. Bozhenkova R.K., Bozhenkova N.L., Shaklsin V.M. Russkiy yazyk i kul’tura rechi: uchebnik. M.: FLINTA: Nauka. 2011. 608 s.
2. Kosarev Yu.A., Li I.V., Ronzhin A.L., Skidanov E.A., Savage J. Obzor metodov ponimaniya rechi I teksta. Trudy. SPIIRAN. 2002. Vyp. 1. T. 2. S. 157-195.
3. Butenko Yu.I., Shostak I.V. Metodicheskie aspekty raspoznavaniya rechi na osnove mnogomernoy statisticheskoy teorii. Neyrokompjutery: razrabotka, primenenie. 2018. № 2. S. 23-33.
4. Voloshin V.G. Komp’uternaya lingvistika. Sumy: Universitetskaya kniga. 2004. 382 s. 
5. Sidnyaev N.I. Khrapov P.V. Neyroseti i neyromatematika: uchebnoe posobie. (pod red. N.I. Sidnyaeva). M.: Izd-vo. MGTU im. N.E. Baumana. 2016. 83 s.
6. Butenko Yu.I. Ispol’zovanie nejronnykh setej dlya obrabotki informatsii po raspoznavaniyu rechi. Mezhdunar. nauch. konf. «Fiziko-matematicheskie problem sozdaniya novoj tekhniki». М. 2014. S. 62.
7. Bondarev V.N., Ade F.G. Iskusstvennyy intellect: Ucheb. Posobie dlya vuzov. Sevastopol’: Izd-vo SevNTU. 2002. 615 s.
8. Frumkina R.M., Vasilevich A.P., Gerganov E.N. Subjektivnye ocenki chastot elementov teksta kak prognoziruuschiy factor. Veroyatnostnoe prognozirovanie v rechi. Sbornik statej. M.: Nauka. 1971. S. 70-93.
9. Frumkina R.M., Vasilevich A.P. Proiznositel’naya trudnost’ bukvosochetanij I ee svyaz’ s porogami zritel’nogo raspoznavaniya. Veroyatnostnoe prognozirovanie v rechi. Sbornik statej. M.: Nauka. 1971. S. 94-134.
10. Sidnyaev N.I. Nejronnaya biovozbudimost’ i postroenie funkcional’nykh skhem iskusstvennogo nejrona. Nejrokomp’utery: razrabotrka, primenenie. 2012. № 6. S. 24-28.
11. Sidnyaev N.I., Shafikova S.E. Nechetkie svedeniya o tochke. Sb. tezisov. nauchno-prakt. konf. «Sovremennye problem matematiki i ee prikladnye aspekty». 2013. S. 61-63.
12. Sidnyaev N.I., Butenko Yu.I., Garazha V.V. Statisticheskaya ocenka associativnoy sily neosmyslennykh bukvosochetanij. Teoreticheskaya i prikladnaya lingvistika. 2019. №5(4). S.107-124.