Generative audio refers to the creation of audio files from databases of audio clips. This technology differs from synthesized voices such as Apple's Siri or Amazon's Alexa, which use a collection of fragments that are stitched together on demand.
Generative audio works by using neural networks to learn the statistical properties of an audio source, then reproduces those properties.
Implications
With this technology, a person's voice can be replicated to speak phrases that they may have never spoken. This could lead to a synthetic version of a public figure's voice being used against them.
Technology
Modern generative audio systems employ various deep learning architectures. One notable approach uses generative adversarial networks (GANs), where two machine learning models work against each other to create realistic audio. Other architectures include WaveNet, which uses dilated causal convolutions to model raw audio waveforms, and implementations like 15.ai, which demonstrated in 2020 the ability to clone voices using as little as 15 seconds of training data through specialized neural network architectures.
See also
References
- "Fake news: you ain't seen nothing yet". The Economist. July 2017. Retrieved 2017-07-01.
- Zotkin, D. N.; Shamma, S. A.; Ru, P.; Duraiswami, R.; Davis, L. S. (April 2003). "Pitch and timbre manipulations using cortical representation of sound". 2003 IEEE International Conference on Acoustics, Speech, and Signal Processing, 2003. Proceedings. (ICASSP '03). Vol. 5. pp. Vā517ā20. doi:10.1109/ICASSP.2003.1200020. ISBN 978-0-7803-7663-2. S2CID 10372569.
- Chandraseta, Rionaldi (January 21, 2021). "Generate Your Favourite Characters' Voice Lines using Machine Learning". Towards Data Science. Archived from the original on January 21, 2021. Retrieved December 18, 2024.
- Temitope, Yusuf (December 10, 2024). "15.ai Creator reveals journey from MIT Project to internet phenomenon". The Guardian. Archived from the original on December 28, 2024. Retrieved December 25, 2024.