![[2602.16721] Speech to Speech Synthesis for Voice Impersonation](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.16721] Speech to Speech Synthesis for Voice Impersonation
Summary
The paper presents the Speech to Speech Synthesis Network (STSSN), a novel model that combines speech recognition and synthesis for effective voice impersonation, outperforming existing generative adversarial models in producing realistic audio samples.
Why It Matters
This research addresses the underexplored area of speech-to-speech processing, which has significant implications for applications in entertainment, accessibility, and security. By advancing voice impersonation technology, it opens avenues for both creative and ethical discussions in AI usage.
Key Takeaways
- Introduction of STSSN model for voice impersonation.
- STSSN outperforms existing generative adversarial models.
- Addresses the gap in speech-to-speech processing research.
- Demonstrates potential applications in various fields.
- Highlights the need for ethical considerations in AI voice technology.
Computer Science > Sound arXiv:2602.16721 (cs) [Submitted on 13 Feb 2026] Title:Speech to Speech Synthesis for Voice Impersonation Authors:Bjorn Johnson, Jared Levy View a PDF of the paper titled Speech to Speech Synthesis for Voice Impersonation, by Bjorn Johnson and Jared Levy View PDF HTML (experimental) Abstract:Numerous models have shown great success in the fields of speech recognition as well as speech synthesis, but models for speech to speech processing have not been heavily explored. We propose Speech to Speech Synthesis Network (STSSN), a model based on current state of the art systems that fuses the two disciplines in order to perform effective speech to speech style transfer for the purpose of voice impersonation. We show that our proposed model is quite powerful, and succeeds in generating realistic audio samples despite a number of drawbacks in its capacity. We benchmark our proposed model by comparing it with a generative adversarial model which accomplishes a similar task, and show that ours produces more convincing results. Comments: Subjects: Sound (cs.SD); Machine Learning (cs.LG); Audio and Speech Processing (eess.AS) Cite as: arXiv:2602.16721 [cs.SD] (or arXiv:2602.16721v1 [cs.SD] for this version) https://doi.org/10.48550/arXiv.2602.16721 Focus to learn more arXiv-issued DOI via DataCite Submission history From: Jared Levy [view email] [v1] Fri, 13 Feb 2026 01:22:25 UTC (1,221 KB) Full-text links: Access Paper: View a PDF of the paper titled Spee...
