First European Conference on Speech Communication and Technology

Paris, France
September 27-29, 1989

Transient Map Method in Stop Consonant Discrimination

Jari Kangas, Teuvo Kohonen

Helsinki University of Technology, Laboratory of Information and Computer Science, Espoo, Finland

Discrimination between the voiceless stop consonants /k,p,t/ is a subproblem in phoneme-based speech recognition systems. Lack of energy during the pronunciation and the fast transient effects at the end of the phoneme make the recognition difficult. A method of so called Phonotopic Maps [2] was studied in order to develop simple and effective solutions for discrimination. In the following studies the method of Transient Maps, a derivative of Phonotopic Maps, was found to be an easy-to-implement and powerful algorithm for real-time speech recognition systems. It contains an automatic learning algorithm that tunes the discrimination elements to detect the differences between the spectra at the end of the stop consonant. Using Transient Maps it is possible to classify correctly 80 to 90 percent of all voiceless stop consonants in our speech recognition system. Thus the recognition accuracy of voiceless stop consonants is comparable to that of the other phonemes.

Full Paper

Bibliographic reference.  Kangas, Jari / Kohonen, Teuvo (1989): "Transient map method in stop consonant discrimination", In EUROSPEECH-1989, 1345-1348.