AI BriefingHurston-1: Open-source TTS model for Southern American and African English accents
·2026.10.25 10:30
Key point
Destined AI releases Hurston-1, a 3B parameter Apache-2.0 model fine-tuned from Maya1 to generate accented English speech with native telephone-band output.
Details
- Hurston-1 is an open-source (Apache-2.0) text-to-speech model by Destined AI, fine-tuned from Maya1, specifically designed to handle Southern American and various African English accents that generic TTS systems often flatten.
- The model introduces NB-TTS (Narrowband TTS), generating 8 kHz G.711 audio natively for telephony systems like Twilio, alongside standard 24 kHz output, trained on approximately 10 hours of real phone calls.
- In blind listening tests, Hurston-1 was preferred over the base Maya1 model 78% of the time for accent authenticity, while maintaining high intelligibility with a mean word error rate of 0.062 on difficult sentences when using the --check flag.
- The model supports voice steering via plain-language descriptions of gender, age, accent, style, and pace, though it does not support voice cloning or consistent voice identity across multiple calls.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Nvidia OpenShell: Secure Runtime for Autonomous AI Agent Fleets