An Evaluation of Language Identification Methods Based on HMMs

Abstract

This paper describes two methods of language identification, both of which are based on HMMs (Hidden Markov Models). Here, we focused on the identification of 10 languages from the OGI Telephone Speech Corpus. In the first method, a fully-structured(ergodic) HMM was trained for each language using text-independent speech samples from many native speakers. The likelihood for each language is calculated for the input speech using this HMM. In the second method, a universal ergodic HMM is trained using all the language data and with it, the most likely state sequence is computed for each language. The state sequence derived is processed and is used in the construction of trigram models for each language. The trigram model was used for modeling the phonotactics for each language. Evaluation on the development test set of the OGI Corpus showed that combining these two methods gave a best performance of 58.5%.