Optimal projections between Gaussian mixture feature spaces for multilingual speech recognition

FREE-DOWNLOAD M Raab, O Schreiner, T Herbig… – Proc. DAGA, Rotterdam, …, 2009
There are two traditional approaches how a system can support the recognition of multiple languages
at the acoustic model level. The first is to run a set of monolingual recognizers in parallel; the
second one is to train a multilingual recognizer for the required set of languages. The first