Abstract
Abstract ASR system is developed for the real time Continuous Speech Recognition of Kannada sententences. The Hidden Markov Model based acoustic models are constructed using Kaldi Tool. The Bigram language model is used for decoding the kannada sentences. The 100 native kannada male and female speakers participated in creation of Kannada speech database and English speech database (South Indian accent). The developed ASR system is evaluated based on the metric Word Error Rate (WER), Wavelet Packet Analysis along with Mel filter bank is utilized to accomplish feature extraction. The proposed features perform slightly better than the existing features like MFCC and PLP under uncontrolled conditions. The proposed features shows slight improvement in performance over baseline features.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have