Title :
Change evaluation of Bayesian detector for dysfluent speech assessment
Author :
Lustyk, Tomá ; Bergl, Petr ; Cmejla, R. ; VokYál, Jan
Author_Institution :
Fac. of Electr. Eng., Czech Tech. Univ. in Prague, Prague, Czech Republic
Abstract :
This paper is focused on a parameter which could be used for automatic and objective assessment of dysfluent speech. The measurement is based on the analysis of read audio recordings of stutterers. The designed parameter analyzes the signal in the spectral domain, which correlates with the subjective assessment of two phoniatrics experts with a coefficient of -0.789. A Bayesian change-point detector (BACD) has been used for spectral analysis. This parameter could become a part of an automatic and objective assessment system.
Keywords :
audio recording; belief networks; correlation methods; speaker recognition; spectral analysis; speech processing; Bayesian change point detector; automatic assessment; dysfluent speech assessment; objective assessment; parameter analysis; phoniatrics expert; read audio recording; spectral analysis; spectral domain; Bayesian methods; Correlation; Databases; Detectors; Educational institutions; Hidden Markov models; Speech;
Conference_Titel :
Applied Electronics (AE), 2011 International Conference on
Conference_Location :
Pilsen
Print_ISBN :
978-1-4577-0315-7
Electronic_ISBN :
1803-7232
