Speech disfluency detection with the correlative method
Abstract
The presented work constitutes a continuation of research on automatic disfluency recognition in utterances by stuttering people. One of the most frequently occurring episodes are syllable repetitions. The repeated fragments have a similar spectral structure, but they differ in their duration times. In order to detect them, correlation of 1/3 octave spectra was applied in connection with the procedures analysing the amplitude-time structure of sound files. The elaborated computer programme allows for recognition of that type of difluency in continuous speech and for exact, graphically illustrated location of the detected episodes in a sound file. It has been verified on the basis of over a hundred 4-second non-fluent utterances. Its functioning has been examined at various border values of the correlation co-efficient and various widths of the time window. Over 70% efficiency of the automatic detection of the episodes has been achieved. The result is comparable to those achieved with the use of the audio monitoring method.
Full Text:
PDFDOI: http://dx.doi.org/10.17951/ai.2005.3.1.132-138
Date of publication: 2015-01-04 00:00:00
Date of submission: 2016-04-27 10:14:19
Statistics
Total abstract view - 503
Downloads (from 2020-06-17) - PDF - 0
Indicators
Refbacks
- There are currently no refbacks.
Copyright (c) 2015 Annales UMCS Sectio AI Informatica
This work is licensed under a Creative Commons Attribution 4.0 International License.