How to Record a Useful Voice Model Sample

Choose material that sounds like you.

Read varied sentences naturally or record clear singing with your own voice. Include comfortable changes in pitch and emphasis, short pauses and the types of sounds you intend to convert. Repeating one phrase or forcing an unfamiliar voice gives the model less useful material.

Keep other speakers, backing music and television out of the recording. For a singing model, record the voice clearly rather than a full mixed song. More minutes of noise are not a substitute for a shorter clean sample.

Keep the room and distance consistent.

Choose a quiet room, silence notifications and place the microphone at a stable distance. Make a short listening check before recording the full sample. If you hear strong echo, move away from reflective surfaces or choose a quieter, less reverberant space.

Start input boost at 0 dB. Speak at your normal level, watch the meter and reduce gain if peaks distort. Raising a quiet recording also raises its background noise; improve the microphone position first.

● ● ●CS Match / Your voice workspaceActual product example

Set gain without clipping →

Review the version you will actually submit.

After browser recording, compare Original with Boosted / enhanced at a comfortable playback volume. Choose the version with clearer speech and fewer artifacts, not simply the louder one. Download WAV if you want a local copy before training.

A training sample must be at least 1 minute and no more than 15 minutes. Uploads are limited to 100 MB. Name the model, confirm that it is your own voice and submit once; wait for ready before attempting song or message conversion.

Open the model workflow →What happens after submission →