Depersonalization of Speech Using Speaker-Specific Transform Based on Long-Term Spectrum
Loading...
Date
2023-12
Authors
ORCID
Advisor
Referee
Mark
Journal Title
Journal ISSN
Volume Title
Publisher
Společnost pro radioelektronické inženýrství
Altmetrics
Abstract
This paper introduces a novel approach for hiding personal information in speech signals. The proposed approach applied a transform warping function, which is obtained from a long-term linear prediction spectrum individually for each speaker. The depersonalized speech was compared with the often used technique based on vocal tract length normalization. The proposed approach performs wider manipulation of fundamental frequency and provides higher intelligibility by 5% in clean speech and by 8% for signal-to-noise ratio 5 dB. It also significantly alters the derived glottal pulses, making them difficult to use for personality analysis. Speech intelligibility index and glottal pulse distortion are new aspects in the field of voice depersonalization.
Description
Citation
Radioengineering. 2023 vol. 32, č. 4, s. 523-530. ISSN 1210-2512
https://www.radioeng.cz/fulltexts/2023/23_04_0523_0530.pdf
https://www.radioeng.cz/fulltexts/2023/23_04_0523_0530.pdf
Document type
Peer-reviewed
Document version
Published version
Date of access to the full text
Language of document
en