Phonetic Engine for Continuous Speech in Malayalam

ABSTRACT

This paper describes the work done towards implementing a phonetic engine (PE) for continuous speech in Malayalam. PE refers to a system that automatically converts input speech to a sequence of phonetic symbols. The PE described in this paper is built for continuous speech in Malayalam, which is speaker-, gender-, and domain-independent. For implementing the PE, a speech database is collected, which consists of speech in (1) read mode, (2) lecture mode, and (3) conversation mode. This data is transcribed using International Phonetic Alphabets. By analyzing the transcription, phonemes are mapped to 40 frequently occurring phonemes, which are then modelled using continuous Hidden Markov Model (HMM). The performance of this PE is evaluated using 1 hour 15 minutes of speech. A Graphical User Interface is developed for the PE to perform real-time recognition of speech.

KEYWORDS:

• Automatic speech recognition
• Hidden Markov Models
• IPA
• MFCC
• Phonetic engine

ACKNOWLEDGMENTS

We thank the Department of Electronics and Information Technology, Government of India, for giving financial support for the consortium project, ‘Prosodically guided phonetic engine for searching speech database in Indian languages’ through which the work described in this paper was carried out. We also thank Miss S. K. Anooja and Mr. Anil P. Antony of Advanced Digital Signal Processing Research Laboratory, Rajiv Gandhi Institute of Technology, Kottayam, for their support towards the work. We thank Prof. B. Yegnanarayana of IIIT Hyderabad (consortium lead of the project) for his immense support.

DISCLOSURE STATEMENT

No potential conflict of interest was reported by the authors.

Additional information

Funding

Department of Electronics and Information Technology, Government of India.

Notes on contributors

Jubin James Thennattil

Jubin James Thennattil received his BTech degree in electronics and communication from Calicut University in 2010 and MTech degree in advanced communication and information from MG University, Kerala. He was working as a research fellow in Advanced Digital Signal Processing Laboratory, Rajiv Gandhi Institute of Technology, Kottayam, during the work mentioned in this paper. He is presently working as an assistant professor at Mar Baselios College of Engineering and Technology, Thiruvananthapuram. His research interest includes spectrum sensing for cognitive radio, statistical signal processing, speech signal processing, and speech recognition.

E-mail: [email protected]; [email protected]

Leena Mary

Leena Mary received her Bachelor's degree from Mangalore University in 1988 and MTech in electronics and communication from Kerala University in 1990. She took her PhD in computer engineering from Indian Institute of Technology, Madras, India, in 2006. She has 24 years of teaching experience. Currently she is working as a professor in electronics and communication engineering at Rajiv Gandhi Institute of Technology, Kottayam, Kerala, India. Her research interests are speech processing, speaker forensics, signal processing, and neural networks. She has published several research papers, which includes a book on Extraction and Representation of Prosody for Speaker, Speech and Language Recognition published by Springer. She has chaired many conferences and is an active reviewer of several journals. She is a member of IEEE and a life member of Indian Society for Technical Education.

E-mail: [email protected]