Language and Dialect Identification

From HLT@INESC-ID

Revision as of 13:02, 29 June 2006 by Imt (talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)

Spoken language identification has been the topic of an MSc thesis [Caseiro, 98] which explored phonotactic features. Our approach was based on an architecture similar to double bigram decoding [Navratil 99]. Our system has been trained and tested on the SPEECHDAT-M database: 9 phonetically rich sentences from 1000 speakers in 6 European languages - English, French, German, Italian, Portuguese and Spanish. The confusion results were in agreement with the known proximity between language pairs. The overall identification rate was 83.4%, and the performance increased, as expected, with the duration of the sentences. The system is easily extendable to other languages, since it only requires audio data in these languages.