5th International Conference on Spoken Language Processing
This paper presents results from a number of flexible vocabulary recognition experiments on the Norwegian SpeechDat(II) database. A common phoneme-based recogniser design procedure is tested on five different tasks, and for five different training sets. Results verify that reasonably accurate recognisers can be built with the database, using standard HMM techniques. They also quantify the importance of training set selection for small and medium vocabulary tasks.
Bibliographic reference. Johansen, Finn Tore (1998): "Phoneme-based recognition for the norwegian speechdat(II) database", In ICSLP-1998, paper 0889.