AUTOMATED SPEECH RECOGNITION SYSTEM OF INDIAN LANGUAGES USING WAVELET TRANSFORMER

Abstract

Spelling correction is the process of finding the appropriate term to substitute for a misspelled word in a text. It is not 
possible for a system intended to fix this problem to know the author's intentions. It should, nevertheless, also find the 
word that the user meant to write. In this work, a recurrent neural network was trained with dictionary phrases and 
used as an oracle. An alternative dictionary word for a misspelled word is provided by this oracle. A character level 
bigram model can be used to create a new query word from a misspelled word. The trained network is additionally 
fed these new query words to acquire more candidate dictionary keywords. The trained network demonstrated a 
satisfactory approach. 
For Indian languages, there aren't many good technological solutions. Wavelet Transformer for Automatic Speech 
Recognition (WTASR), the proposed technology, addresses this vacuum. Transforming speech impulses into 
equivalent text representations is the system's primary goal. Because of this, it can be used for a variety of purposes, 
such as chatbots, assistive technology, and voice-activated instructions. In an effort to bridge the technological divide 
in Indian languages, the suggested Wavelet Transformer for Automatic voice Recognition (WTASR) offers a practical 
way to translate speech signals into equivalent text.