Implementation of Bangla Speech Recognition in Voice Input Speech Output (VISO) Calculator

摘要

Automatic Speech-recognition (ASR) is an engaging way for human-computer interaction. It is a process of deriving linguistic information from acoustic signals while removing noise signal associated with it. The speaker-independent Voice Input Speech Output (VISO) calculator application is a promising implementation of speech-recognition technology, which is designed to recognize numbers and symbols to perform mathematical operations. But none of such applications provide support for Bangla language yet. Thus, in this paper, we have designed a VISO calculator application that can recognize both isolated and continuous speech in Bangla language and derive mathematical expression. The expression is then calculated and the auditory output is pronounced in Bangla. The task of speech-to-text (STT) conversion is performed by Hidden Markov models (HMM) in CMU Sphinx, while the result is converted to speech using Android Text-to-Speech (TTS) API. The application is able to recognize Bangla real numbers such as , (3) etc. and several mathematical symbols such as , (.) etc. This voice-based application is helpful for people having physical disabilities such as congenital disorder or visual impairment and people with less mathematical knowledge, who use Bangla as their native language.