Show HN: Hurston-1 open source TTS that speak Southern American and Afro-English (huggingface.co)

Destined AI has launched Hurston-1, an open-source text-to-speech (TTS) model designed to accurately emulate Southern American and Afro-English accents. Named after anthropologist Zora Neale Hurston, this innovative model employs Narrowband TTS (NB-TTS), specifically optimized for phone use, where the majority of customer service interactions occur. Hurston-1 is fine-tuned from the Maya1 model, allowing it to generate realistic speech from real calls, making it more effective in delivering what users genuinely experience over phone lines. The model supports 8 kHz G.711 audio suitable for telephony systems, providing a direct response to a long-standing gap in the representation of diverse English dialects in AI systems. Significantly, Hurston-1 addresses the inadequate performance of existing TTS systems in accurately capturing regional accents, which historically get flattened into generic American English. It was trained using a diverse dataset of 1,800+ hours, incorporating recordings from consented contributors across the Southern U.S. and various African regions. The model showcases robust accent classification and intelligibility, outperforming prior TTS iterations in capturing the nuances of African and Southern American speech. Applications span from enhancing customer service AI agents to educational tools for accent training, emphasizing the need for inclusivity in AI voice technology and reflecting community voices in synthetic speech synthesis.