Hurston-1

Narrowband TTS (NB-TTS): accented text-to-speech built for the phone. Hurston-1 speaks Southern American, Nigerian, West African, East African and Southern African English, and its phone mode delivers exactly what a caller hears. Describe a voice in plain words; Hurston-1 speaks it.

▶ Listen Get the model License the data

78%of blind comparisons chose Hurston-1 over the base model

0.84–0.98accent-classifier score for African voices (base model: 0.60–0.87)

1,800+ hof consented, paid contributor speech recorded since 2021. Hurston-1 was trained on a ~136 h slice.

Hear the South

A few lines the way people actually talk around Atlanta.

Guess the accent

Ten AI voices, nine accents. Can you tell Lagos from Accra, or Atlanta from the rest of America? Answers are anonymous and help us improve the next model.

Built for the phone

Most customer service still happens on a phone line, even when an AI agent answers. Hurston-1 learned in part from ~10 hours of real calls between our contributors and AI voice agents, and phone mode outputs the 300–3,400 Hz telephone band as 8 kHz G.711, ready for Twilio and SIP. Hear the same take both ways.

Every accent

Same model, one description each. Every voice is generated; none is a real person.

“The South got something to say.”

In 1995 André 3000 said it on stage at the Source Awards, to a crowd that was booing OutKast. We think that line still fits.

Hurston-1 was built in Atlanta by a small, independent team. It is named for Zora Neale Hurston, who traveled the South recording how Black Southerners actually talked, at a time when few thought those voices were worth keeping. The name is a tribute and implies no endorsement by her estate.

Probability judged African	Base	Hurston-1
Southern African	0.60	0.95
West African	0.60	0.84
East African	0.81	0.98
Nigerian	0.82	0.94
American (should be low)	0.12	0.02

Accent classifier trained on real speakers; base Maya1 prompted in its own documented style. Full evaluation on the model card.

Try it yourself

Runs on one GPU. The voice description is plain text: gender, age, accent, style, pace.

pip install torch transformers snac soundfile scipy speechbrain
huggingface-cli download destined/hurston-1 --local-dir hurston-1 && cd hurston-1

python hurston.py --check --accent "Southern American accent" --gender Male \
    --text "The South got somethin to say." --out south.wav

# as a caller hears it: 8 kHz G.711 mu-law
python hurston.py --check --phone --accent "Nigerian English accent" --gender Female \
    --text "Thanks for calling, how can I help you today?" --out call.wav

What's next

Hurston-1 is the first step. Next: telephony-native TTS that learns to talk like a person on a call (turn-taking, “mm-hm”s, call pacing, a call style distinct from reading), and consistent voices that stay the same across every call. Follow Destined AI on Hugging Face.