Since it is by no means easy to achieve good phoneme recognition rate for noisy telephone speech, N-gram built upon recognized phoneme labels was evaluated and was found to be more effective than the N-gram built upon original attached phoneme labels for language identification. The performance of mixed phoneme recognizer, in which both language-dependent and language-independent phonemes were included, was also evaluated. Results showed the performance was better than that using parallel language-dependent phoneme recognizers in which bias existed due to different numbers of phonemes among languages.
Bibliographic reference. Kwan, HingKeung / Hirose, Keikichi (1995): "Recognized phoneme-based N-gram modeling in automatic language identification", In EUROSPEECH-1995, 1367-1370.