Entropy shows: is it real speech, or a clone?
DOI:
https://doi.org/10.36505/ExLing-2024/15/0003/000628Keywords:
entropy, cloned speech, forensics, formants, AIAbstract
This exploratory study compares the conditional entropy formant values of naturally spoken words with those of cloned vowels. The study hypothesized that the cloned vowel formants — F1, F2, and F3 — would display measurable and distinct differences when compared to the formants found in natural speech. The findings confirmed that there are notable variations in the entropy values of Artificial Intelligence (AI)-generated cloned vowels. These differences may be valuable in forensic investigations, particularly in distinguishing between authentic human speech and AI-generated voice imitations.
References
Botinis, A., Goertz, G., Kontostavlaki, A., Anderson, T. 2023. Vowel discrimination of American English. ExLing 2023 Athens: Proceedings 14th International Conference of Experimental Linguistics, October 13-16. Athens, Greece.
The MathWorks Inc. 2023. MATLAB version: 9.13.0 (R2022b), Natick, Massachusetts: The MathWorks Inc. https://www.mathworks.com.
Goodfellow, I., Bengio, Y., Courville, A. 2016. Deep Learning. Cambridge, Massachusetts: The MIT Press.
Haglund, J.F. Jeppsson, H. Strömdahl. 2010. Different Senses of Entropy – Implications for Education. Entropy 12, 490-515. Doi: 10.3390/e12030490.
Peng, H. 2022. Mutual Information computation. (www.mathworks.com/matlabcentral/fileexchange/14888-mutual-information-com).
Downloads
Published
Issue
Section
License
Articles are published under the Creative Commons Attribution 4.0 International License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are properly credited.
