Creating a Grammar-Based Speech Recognition Parser for Mexican Spanish Using HTK, Compatible with CMU Sphinx-III System

Abstract

In this paper, we present the creation of a novel language model for automatic speech recognition based on HTK but compatible with the CMU Sphinx-III speech recognition system. This recognizer works with user defined grammars in the HTK format for speaker dependent recognition in Mexican Spanish. Input files are based on the Sphinx format, so you can use them with no modification in both systems. We provide with the recognizer a "live decode" module based on the Japanese speech recognition system called Julius for live recognition. We will also explain the phonetic alphabet chosen for the system and we will explore some of the new resources that are now available for the Mexican Spanish language. This software is available for free use and it is compatible with the most common operating systems (Windows, GNU-Linux and MAC OS).