In this paper, we have discussed about the Vowel Onset Point (VOP) for the Hindi language and its significance in the speech recognition. We have defined the vowel onset point and how it can be calculated. Alphabets in Hindi language are the combination of the vowel and consonant part. In Hindi, we cannot pronounce a consonant without a vowel. There is a very small region between consonant and vowel where transition happens from consonant to vowel. We have used characteristics of the sound files to get the vowel onset point. To calculate Vowel Onset Point, we have applied filtration process, and after that, we can use energy of the signal and different formants combined with epoch interval and Itakura distance. Filtered energy and filtered formants can be used as cues for accurately detecting VOP within the range of +/-30 ms. In order to further increase the effectiveness of the proposed method, we have used Recurrent Neural Network variants to detect VOP which uses speech features and reference point calculated by filtered formants.
展开▼