Echo voices and Baobab national models — Africa-first language intelligence.
Explore BaobabText-to-speech
Region-aware speech, Echo Instant Clone, and extract/isolate — ready for videos, songs, and agents.
Lugemi Voice is the sound layer of the platform. Generate with own:* voices, enroll Instant clones from recorded samples (consent + review), extract tracks from uploads, and reuse approved clone:{id} in Studio, Dubbing, and developer tools. Native-accent pickers expose country, ethnic/community context, gender, and tone/emotion style — Africa-first. Realtime means instant enroll — not live training theater.
Press Play to hear this script in Ama · Ghanaian English (en-GH).
Try region voices
Voice Cloning — product answers
Can people clone voice in realtime and use it in videos / songs?
Echo Instant Clone ships: short consent sample → abuse review → speak with clone:{id}. That is the closest honest realtime product — not live model training. Use approved clones in Voice Studio, Dubbing, Chat Studio, and MCP/CLI for video narration or song vocals (DAW mix stays yours).
Can they upload a recorded voice?
Yes. Voice Studio accepts 1–5 recorded samples or in-browser mic capture with consent attestation and notes (Pro + review).
Can they extract voice from uploaded files?
Yes for audio-track extract (browser) and Lugemi isolate (energy VAD). Entertainment-grade neural stem separation is not offered on this surface.
How teams ship with Lugemi Voice: interactive demos above, step guides below, and the same /v1 surface for APIs and SDKs.
Use case guide
Generate a sample, enroll a consented clone, then drop clone:{id} into a project.
API & SDK
First-party synthesize and clone endpoints with workspace metering.
Produce narration and agent replies with Africa-first accents and honest availability.
own:* catalog voices carry Ghanaian English, Asante/Akuapem Twi, Ga, Ewe, Kenya Swahili, Addis Amharic, Cairene Arabic, Senegalese French, and more — with male/female and tone tags for builders. Default speaking-agent path is Ghana (Accra). Asante Twi neural corridors ship where live; otherwise Play remaps those demos to a verified Ghanaian English path.
Upload recorded voice: yes. Extract from files: yes (track extract + energy VAD isolate; neural stems not on this surface). Realtime: instant clone after consent, then watermarked speak — not live model training.
