An automatic intonation tone contour labelling and classification algorithm

Automatic recognition and classification of intonation contours in natural speech has the potential to improve the performance and useability of speech recognition systems, particularly those employed in automated speech response applications. The role of intonation in signalling meaning in natural English speech is well understood by linguists, and has been incorporated into the functional systemic grammar model of language. In this paper we describe the first stage of the development of a rule-based algorithm for intonation contour labelling and recognition which takes as its data input a stylised piecewise-linear approximation to the speech pitch contour. A key feature of the algorithm is the use of perceptually-based weighting of the segmented contour. Our results show good correspondence between the results of the automatic contour matching process and that of an expert human listener for a wide range of contours.<
>