Voice dataset · Asante Twi

Nwene

"Nwene" — to weave. Record yourself speaking natural Asante Twi, prompt by prompt or free-flowing, and weave it into a dataset ready for training a voice model that sounds like you.

0 recorded · 0:00 total
GREETING

Maakye

Good morning

1 / 36
0:00

Requesting microphone access…

Recorded takes

0 saved

Nothing recorded yet — your first take will show up here.

Record a few takes, then export your dataset.
Recording tips: use a quiet room, keep a consistent distance from the mic, and speak at your natural pace — don't over-enunciate for the machine. Aim for variety: questions, statements, and exclamations all carry different tone patterns in Twi, and the model needs to hear all of them. More data helps more than anything else here — even 1–3 hours of clean audio makes a real difference over a few minutes.

What's in the export: a .zip with one audio file per take plus a manifest.csv and manifest.json pairing each file with its transcript and category — the standard format most TTS training pipelines expect. Audio is captured as .webm; convert to .wav with a tool like ffmpeg if your training pipeline requires it.