Accent detection of Telugu speech using prosodic and formant features

Speech automation is becoming more popular in the recent times. Speech recognition systems are increasing day by day. Earlier the speech recognition systems were developed for English language. Now these systems are being developed for many other languages. Many languages in the globe have different speaking styles or accents. The speech recognition systems may not recognize speeches with accent other than the system are trained. So it is important in the speech to text conversion systems to convert the accented speech in to text. Telugu is a language of southern part of India, has mainly three different accents namely Coastal Andhra, Rayalaseema and Telangana, in which the stress is different for the same word in these accents. In this work, text dependent speeches from Coastal Andhra, Rayalaseema, Telangana accents have been collected. Prosodie and formant features extracted from speech are used for discriminating the accents. Prosodie features are represented by durations of syllables, pitch and energy contours. These features are used to recognize the accent of the speaker using Nearest Neighborhood Classifier. The best recognition Accuracy using this model obtained 72%.