Acoustic Feature Extraction and Isolated Word Recognition of Speech Signal Using HMM for Different Dialects

In this paper improved the recognition rate of isolated words using the hidden Markov model (HMM) with improved accuracy at all frequency ranges of the audio voice. In English and Punjabi, two different dialects of isolated words are recorded using a microphone and audacity tool. These words cover all the phonemes of respective dialects. This work is mainly divided into two sections: feature extraction and recognition. In these techniques, Mel-frequency cepstral coefficients (MFCC) have been used to extract features mathematically. The second part, described the Markov model and HMM with examples for modeling and recognizing the isolated words. Based on the K-mean algorithm, all the samples are clustered. Where the Gaussian mixture contains weight modeling, mean, and variance parameters. The K-folded method used here for training and testing each signal uses the forward and Baum-Welch algorithms. Finally, it gives a minimum misclassification rate from the N number of different hidden states and different values of a number of the feature vector in each frame D. The novelty of this article is to recognize the isolated word using HMM for English and Punjabi Dialect and made a comparison between them. For the computation, five folded cross-validation is used. The misclassification rate of vocabulary is computed using HMM and acoustic models. The database is created using English and Punjabi phonemes in a real-time environment. The training and testing are performed to validate the robustness of the system. The experimental results have shown the system performance to be 93% for English vocabulary and 96% for Punjabi vocabulary.