Dysphonia Detected by Pattern Recognition of Spectral Composition The vowel [a:] in a test word, judged normal or dysphonic, was examined with the Self-Organizing Map, the artificial neural network algorithm of Kohonen. The algorithm produces two-dimensional representations (maps) of speech. Input to the acoustic maps consisted of 15-component spectral vectors calculated at 9.83-msec intervals from short-time power spectra. ... Research Article
Research Article  |   April 01, 1992
Dysphonia Detected by Pattern Recognition of Spectral Composition
 
Author Affiliations & Notes
  • Lea Leinonen
    Department of Physiology University of Helsinki
  • Jari Kangas
    Laboratory of Information and Computer Science Helsinki University of Technology
  • Kari Torkkola
    Laboratory of Information and Computer Science Helsinki University of Technology
  • Anja Juvas
    Department of Phoniatry Helsinki University Central Hospital
Article Information
Speech, Voice & Prosodic Disorders / Voice Disorders / Speech, Voice & Prosody / Speech / Research Articles
Research Article   |   April 01, 1992
Dysphonia Detected by Pattern Recognition of Spectral Composition
Journal of Speech, Language, and Hearing Research, April 1992, Vol. 35, 287-295. doi:10.1044/jshr.3502.287
History: Received December 17, 1990 , Accepted August 7, 1991
 
Journal of Speech, Language, and Hearing Research, April 1992, Vol. 35, 287-295. doi:10.1044/jshr.3502.287
History: Received December 17, 1990; Accepted August 7, 1991

The vowel [a:] in a test word, judged normal or dysphonic, was examined with the Self-Organizing Map, the artificial neural network algorithm of Kohonen. The algorithm produces two-dimensional representations (maps) of speech. Input to the acoustic maps consisted of 15-component spectral vectors calculated at 9.83-msec intervals from short-time power spectra. The male and female maps were first calculated from the speech of healthy subjects and then the [a:] samples (15 successive spectral vectors) were examined on the maps. The dysphonic voices deviated from the norm both in the composition of the short-time power spectra (characterized by the dislocation of the trajectory pattern on the map) and in the stability of the spectrum during the performance (characterized by the pattern of the trajectory on the map). Rough voices were distinguished from breathy ones by their patterns on the map. With the limited speech material, an index for the degree of pathology could not be determined. A self-organized acoustic map provides an on-line visual representation of voice and speech in an easily understandable form. The method is thus suitable not only for diagnostic but also for educational and therapeutic purposes.

Acknowledgments
This study was carried out in Teuvo Kohonen’s laboratory. We are grateful for his support of a study with more practical than theoretical implications. We wish to thank Matti Lehtihalmes, Anna-Maija Korpijaakko-Huuhka, and Reijo Aulanko, from the Department of Phonetics, University of Helsinki, for advice and practical help during the study. Maija Laakso from the Department of Physiology, University of Helsinki, advised us in the statistical testing. We are also grateful to Heikki Rihkanen, Department of Otorhinolaryngology, Helsinki University Central Hospital, for the critical comments on the manuscript. We also wish to thank llkka Linnankoski for aid in clarifying the linguistic expression. The study was financially supported by Academy of Finland.
Order a Subscription
Pay Per View
Entire Journal of Speech, Language, and Hearing Research content & archive
24-hour access
This Article
24-hour access