Artificial intelligence

Abstract: This paper describes the development of experimental systems of speech signal to text conversion based on words and sub-words. Main attention is paid to selecting of training set for estimation of the parameters of acoustic recognition models. Particularly, the following options are considered: acoustic model based only on continuous speech, a model that integrates continuous speech and isolated words, a model that ignores stress vowels, and a model that takes into account only stress vowels “y” and “e”. The estimation of acoustic model parameters is based on mono-speaker speech corpus. The factors compensating the inconsistency of acoustic and linguistic component model scales are analyzed and their values are explored. The results of experimental research are discussed.

Keywords: sub-word, syllable, speech recognition, training and test sets, continuous speech, isolated words.

References: