Automatic language identification using syllabic spectral features

Automatically identifying a language from just the acoustics is a challenging problem. Speaker differences are usually greater than language differences. The study has developed a text-independent system that is capable of performing both speaker and language identification. The system utilized different feature sets to observe changes in recognition performance to identify which set of features is suitable for language identification. Through these experimental results, the spectral features at the syllabic level have proven to be reliable for distinguishing languages. Performance on a five language database has exceeded 95% identification accuracy. Two other telephone-speech databases were also tested