Automatic Speech Recognition System for Real Time Applications

Speech is human’s most efficient communication mode. Beyond its efficiency, humans are comfortable and familiar with speech. Other modalities require more concentration, restrict movement and cause body strain due to unnatural positions. This need brings the development of a speech to text conversion. In spoken language, syllables are often considered as the phonological "building blocks" of words. Depending on the language and the sounds used, a phoneme may be written consistently with one letter; however, there are many exceptions to this rule. The range of the possible applications is wide and includes: voice-controlled appliances, fully featured speech-to-text conversion, automation of operator-assisted services, and voice recognition aids for the handicapped. This project implements a removal of additive noise and conversion of speech to text form. Spectral substraction is used to remove the noise present in speech. Next is the segmentation process done with the help of group delay algorithm. Recognition plays a major role in speech to text conversion. Letter recognition is achieved with simple Gaussian Mixture Model (GMM). Word recognition is a challenging scenario for researchers and is extracted by HMM with more accuracy main application of the framework is hand free data entry. Mobile and medical environment also use this speech to text conversion.