Robust Syllable Segmentation Of Continuous Speech Using Neural Networks

Automatic blind syllable segmentation for continuous speech

“…The output of each algorithm was a set of detected syllable boundaries, Y. These were compared with the hand labelled reference syllable boundaries X to count the number of correct detections, insertion errors and deletion errors based on equations (4), (5) and (6) respectively. A detected boundary y i within a This score eliminates boundaries which precede onsets that peak at a low velocity (less than b min ).…”

“…This has lead to the development of more sophisticated techniques such as that by [5] which uses self organising maps and techniques by [6,7] which both use complex models of the relationships between the linguistic and speech processes involved. These techniques appear to yield improvements in segmentation have the disadvantage of being difficult to implement and none have been accepted as an algorithm of choice for speech researchers.…”