Creating a Mexican Spanish Version of the CMU Sphinx-III Speech Recognition System

摘要

In this paper we present the creation of a Mexican Spanish version of the CMU Sphinx-III speech recognition system. We trained acoustic and N-gram language models with a phonetic set of 23 phonemes. Our speech data for training and testing was collected from an auto-attendant system under telephone environments. We present experiments with different language models. Our best result scored an overall error rate of 6.32%. Using this version is now possible to develop speech applications for Spanish speaking communities. This version of the CMU Sphinx system is freely available for non-commercial use under request.