Effect of sleepiness in the voice on speaker recognition performance

The issue of the influence of speaker state on voice recognition has been analysed mainly in relation to forensics and biometric security systems. Sleepiness in the voice is a rather under-researched problem, and the few works in this area focus almost exclusively on the recognition of sleepiness rather than on its influence on the change of the speaker's voice characteristics. This paper discusses the issue of the influence of the speaker's state on voice recognition, describes the acquisition method of the acoustic database of voice drowsiness recordings used in the tests. It also discusses the subjective sleepiness scales used in the study and presents the results of the influence of sleepiness on the effectiveness of automatic speaker recognition based on a classical system using the Mel-Frequency Cepstral Coefficients parameterisation and the Gaussian Mixture Models classification.