Research Cell: An International Journal of Engineering Sciences

Open Access Open Access  Restricted Access Subscription Access

Malayalam Word Identification for Speech Recognition System


Affiliations
1 Indian Institute of Information Technology and Management (IIITM-K), Kerala, India
 

Automatic Speech Recognition (ASR) Systems have long been a goal of artificial intelligence researchers. The lack of state-of-the art ASR System has been a major hindrance due to its complexity in reproducing in Computer. Hidden Markov Models (HMMs) are used heavily in most current speech recognition systems for both phoneme and syllable based approach. In this paper, we also propose to use HMM model, but based on energy measure and Mel Frequency Cepstral Coefficient (MFCC) to determine the syllable based segmentation and features of power spectrum of speech signal. The system was trained with utterances of 3 male and 2 female speakers and the database included 40 utterances. The training and testing was done with bi-syllable words and the implementation of the system has been done using Hidden Markov Model Toolkit (HTK).
User
Notifications
Font Size

Abstract Views: 196

PDF Views: 3




  • Malayalam Word Identification for Speech Recognition System

Abstract Views: 196  |  PDF Views: 3

Authors

Maya Moneykumar
Indian Institute of Information Technology and Management (IIITM-K), Kerala, India
Elizabeth Sherly
Indian Institute of Information Technology and Management (IIITM-K), Kerala, India
Win Sam Varghese
Indian Institute of Information Technology and Management (IIITM-K), Kerala, India

Abstract


Automatic Speech Recognition (ASR) Systems have long been a goal of artificial intelligence researchers. The lack of state-of-the art ASR System has been a major hindrance due to its complexity in reproducing in Computer. Hidden Markov Models (HMMs) are used heavily in most current speech recognition systems for both phoneme and syllable based approach. In this paper, we also propose to use HMM model, but based on energy measure and Mel Frequency Cepstral Coefficient (MFCC) to determine the syllable based segmentation and features of power spectrum of speech signal. The system was trained with utterances of 3 male and 2 female speakers and the database included 40 utterances. The training and testing was done with bi-syllable words and the implementation of the system has been done using Hidden Markov Model Toolkit (HTK).