← All models
Text to speech model · Liberia

mahkeh-voice v2.3

mahkeh-voice reads English, Koloqua and Kpelle aloud with Liberian accents, so people can hear the words the way they are spoken at home.

English + Koloqua + Kpelle5 voicesUp to 1,500 characters
Listen in the Playground
What it does
Written text→Spoken audio

Speech generation with Liberian accents.

Release notes

Current: v2.3

mahkeh-voice v2.3: Five Liberian voices, now with Kpelle

CurrentNovember 2026

What’s new

  • Five named voices: Varney (Calm), Musu (Calm), Sackie (Koloqua guy), Pa Tarnue (Moderate) and Pa Kollie (Kpelle man).
  • Reads Kpelle as well as English and Koloqua; Pa Kollie reads Kpelle by default.
  • Long text is no longer cut off: a sentence that stops early is read again.

Known limits

Delivery can still sound flat, and Kpelle pronunciation is still improving as we improve the model with more Kpelle speakers. Some Kpelle translations may not be very accurate.

mahkeh-voice v1.8: Stronger Liberian accent

October 2026

What’s new

  • Bigger base model for a stronger Liberian accent.
  • Five voices, including the new Liberian woman, moderate voice.

Known limits

Speech delivery is still somewhat flat, and some pronunciations may drift toward a more general West African accent. Koloqua voice quality also needs improvement, and we are continuing to train the model to produce more natural, authentic Liberian Koloqua speech.

mahkeh-voice v1.7: Liberian text to speech

October 2026

What’s new

  • Four voices: Liberian man, calm; Liberian man, tough; Liberian girl, calm; Liberian man, moderate (Koloqua).
  • Trained on audio exactly as recorded, for more natural voices than v1.6.
  • Fewer long pauses and no runaway takes.

Benchmark

Listening tests by a native speaker on fixed test lines in English and Koloqua.