Speaker Accent and Isolated Kannada Word Recognition

Algorithm is designed for isolated Kannada word recognition of five districts Kannada speakers’ accent. Isolated Kannada words recognition is designed using the syllables, Baum-Welch algorithm and Normal fit method. The novelty of proposed method is in recognition of five district Kannada speaker accents as well as spoken words. Our model is compared with baseline Hidden Markov Model (HMM) and Gaussian Mixture Model (GMM). Our model is tested for both noisy (little) and noiseless signal. For experiment totally 7056 signals are used for training and 3528 signals are used for testing. The experiment showed that average WRR for known accent speaker is 95.56% and of unknown accent speaker is 90.86% and average recognition of Kannada speaker accents is 82%.

Author(s): Hemakumar G, Punitha P

Abstract | PDF

Share This Article