ITRW on Non-Linear Speech Processing (NOLISP 05) Barcelona, Spain April 19-22, 2005

Segment Boundaries in Low Latency Phonetic Recognition

Giampiero Salvi

KTH, Royal Institute of Technology, TMH dept. for Speech, Music and Hearing, Stockholm, Sweden

This study analyses how the reduction of the look-ahead length of a two pass phonetic decoder in uences the alignment of the segment boundaries. It is shown how the optimization of some tuning parameters, such as the insertion penalty, is dependent on the look-ahead length. It is also suggested that the insertion penalty be dynamically adjusted to some measure of similarity of the phonetic segments following each other.

Full Paper

Bibliographic reference.  Salvi, Giampiero (2005): "Segment boundaries in low latency phonetic recognition", In NOLISP-2005, 145-150.