On-line recognition of vocalised Pitman shorthand outlines

The obvious method of achieving simultaneous transcription of speech would be by using a speech recogniser. Unfortunately, the difficulties involved in the recognition of unconstrained, unlimited vocabulary connected speech are daunting and will not be solved for many years. Therefore, alternative methods have been investigated which require a human intermediary to carry out the computationally difficult tasks of isolating the speakers voice from other background sounds, separating words in the speech, ignoring the redundant speech utterances and simultaneously encoding the essential speech components of each word via some other medium such as keyboard or handwriting. The authors describe an approach for the segmentation and classification of the consonant part of a vocalised Pitman outline into its phonetic features involves three stages. Each segment is examined to determine whether it is long enough to be a primitive. Once the segmentation has been completed the features are classified using a fuzzy technique which classifies each feature into categories in terms of the degree of membership. Finally, decision making is achieved using the maximum likelihood principle