Open Access Open Access  Restricted Access Subscription or Fee Access

Developing Telugu Speech Recognition System using Sphinx-4

P. Jayaprakash, K. Venkataramana, E. Prakashbabu

Abstract


Speech is the main communication medium in Human beings. Speech recognition has many applications. It can be used to automate many tasks that previously required hands-on human interaction. Many speech recognition systems have been proposed and developed. Sphinx4 is one such system developed in Java for recognizing English. In this paper how speech recognition is done is presented briefly and then sphinx4 architecture is explained. This paper aims at installing sphinx4 and testing it. Sphinx4 was used to develop applications for Telugu. This is done by identifying the language dependent areas of the sphinx4 and their interactions with other parts of the system and then modifying these areas so that it works for Telugu. Static Dictionary for Telugu words is developed rather than using online ‘lmtool’ of CMU. Sphinx4 was adapted to train and decode for recognizing isolated Telugu words. Continuing in a similar manner, a speech recognition system for Telugu can be developed.

Keywords


Melfrequency Cepstral Coefficients, Hidden Markov Models, Language Model, N-Grams

Full Text:

PDF

References


W.A.Lea, “Trends in Speech Recognition”, Prentice Hall, 1980.

http://www-2.cs.cmu.edu/~robust/Tutorial

Modelling Word Duration for Better Speech Recognition by Venkata Ramana Rao Gadde SRI International Menlo Park, USA

http://www.cis.hut.fi/jpylkkon/pylkonen04norsig.pdf

http://www.utdallas.edu/~loizou/cimplants/hillen.pdf

http://www.mmk.e-technik.tu-muenchen.de/pb/ps/00pfa1.pdf

.http://research.microsoft.com/srg/papers/1999-richardsoneurospeech.pdf


Refbacks

  • There are currently no refbacks.