Atlanta-based Destined AI has open-sourced Hurston-1, a narrowband text-to-speech model (Apache-2.0) optimized for telephone-quality audio (8 kHz G.711) and full 24 kHz output. The model excels at Southern American, Nigerian, West African, East African, and Southern African English accents — voices often flattened by mainstream TTS systems. Built as a fine-tune of Maya1 using 136 hours from a 1,800-hour consented dataset of over 20,000 contributors collected since 2021, Hurston-1 is designed for telephony integration with Twilio and similar platforms.
The founder reports $500K+ revenue from enterprise customers including Amazon and Salesforce. Limitations include inconsistent voice identity across prompts and no voice-cloning support. Named for anthropologist Zora Neale Hurston, the model aims to bring underrepresented voices into AI.
The team invites feedback via Hugging Face and collaboration with others in the space.
Source: Hacker News · Summarized by HeadlinesBriefing