Artificial intelligence

Scientific journal

ISSN 2710-1673

ONLINE: ISSN 2710-1681

Select your language


On the Problem of Continuous Speech Recognition

Shelepov V.1, Nicenko A.1
1 Institute of Artificial Intelligence MES of Ukraine and NAS of Ukraine

Full text (PDF)

UDC: 004.934.1’1
Publication Language: Russian
Stuc. intelekt. 2012; 17(4):272-282

Abstract: Author’s recognition of continuous speech method is describing in the article. Phrase is consisting of words of given vocabulary. The method is based on using author’s mechanism of segmentation speech signal and algorithm of searching the first word, which we name minimal DTW-distant principle. We use earlier proposed by authors procedure of construction patterns with the help of diphone base and some modification of method DTW which gives win in speed and memory. Practical realization of method demands of more precise determination of speech boundaries and enlargement quantity rules of automatic transcription system. This is the subject of two first sections of the article.

Keywords: beginning and end of speech, transcription system, continuous speech, segmentation, diphone, DTW-algorithm, the first word separation, minimal DTW-distant principle

References:

View full text (PDF)