Automatic spoken language identification is the task of identifying the language from a short duration of a speech signal. One of the important language identification cues is the differences in phoneme frequencies among different languages. Considering this, we develop a pairwise multiple codebook approach to language identification. This system is compared with the traditional single codebook per language system. Traditional VQ based language models are generally preferred since they do not explicitly require language models. The evaluation with Oregon Graduate Institute Multi-Language Telephone Speech Corpus shows that the multiple codebook system improves the performance by almost 7%.
Cite as: Nagarajan, T., Murthy, H.A. (2003) A pairwise multiple codebook approach to implicit language identification. Proc. Workshop on Spoken Language Processing, 101-108
@inproceedings{nagarajan03_wslp,
author={T. Nagarajan and Hema A. Murthy},
title={{A pairwise multiple codebook approach to implicit language identification}},
year=2003,
booktitle={Proc. Workshop on Spoken Language Processing},
pages={101--108}
}