Speech Recognition System for Different Kannada Dialects

In this paper discuss on pronunciation variations occurs in different Kannada dialects, designing language model, building acoustic models, finally recognition of Kannada dialect speech. Algorithm designed for recognition of isolated Kannada word and continuous Kannada speech made by different dialects speakers. The novelty of algorithm is in handling multiple Kannada dialects speaker’s speech recorded by mini-microphone, headphone and cell phones. Robustness of the algorithm in handling different Kannada dialects speech and handling little noisy waves. Here speech waves recorded at natural environment. Here classification of speech models based on speaker’s dialects and inside the dialects sub classes designed according to acoustic features. During recognition, breadth first matching technique and then inside that dialect class depth first matching techniques implemented. Here speech recognition designed using MFCC and coefficients of real cepstrum features and compared the performance. In these experiment real cepstrum coefficients, features produced better recognition rate while dealing with multiple dialects of same language. All computations made using mat lab.