Syllable Based Deep Learning for Multi-Dialect Speech Recognition

  Rohan Gupta,  Rohan Kumar

As a by-product of India’s linguistically diverse culture, it is estimated that dialects and accents change noticeably every 100km radius. Considering India’s burgeoning technological industry as well as our own project to aid the visually impaired, in this paper, we explore methods to effectively apply Automatic Speech Recognition techniques across multiple dialects of Indian languages. We make use of a novel, syllable-based Hidden Markov Model in conjunction with a multi-layered, modified Artificial Neural Network to decode and analyze speech, specifically formal commands. Our research suggests that syllable-based speech recognition presents a lucrative mode of sequential pattern recognition, a consequence of linguistic properties within dialects and phonology of Indian languages.

Keywords- Deep Learning, Automatic Speech Recognition, Syllabification, Multi-Dialect

Unique Identification Number - IJEDR1904106
Page Number(s) - 636-641
Pubished in - Volume 7 | Issue 4 | December 2019
DOI (Digital Object Identifier) -    http://doi.one/10.1729/Journal.22906
Publisher - IJEDR (ISSN - 2321-9939)

  Rohan Gupta,  Rohan Kumar,   "Syllable Based Deep Learning for Multi-Dialect Speech Recognition", International Journal of Engineering Development and Research (IJEDR), ISSN:2321-9939, Volume.7, Issue 4, pp.636-641, December 2019, Available at :http://www.ijedr.org/papers/IJEDR1904106.pdf