The four things that matter
- Background noise: hum, hiss, traffic and fans. It is copied into the clone.
- Room echo (reverb): makes the clone sound distant. Record close to the mic in a soft room.
- Levels: too quiet adds hiss, too loud distorts. Speak at a normal volume and avoid clipping. The Studio normalises the level automatically.
- Single speaker: other voices, music or effects confuse the model.
Clean vs noisy reference: listen
Formats and sample rates
You can upload WAV, MP3, M4A or OGG. Everything is converted to mono 24 kHz internally, so recording at 44.1 or 48 kHz is fine. Prefer WAV or high-bitrate MP3 (192 kbps or more): low-bitrate files add artefacts.
The output
Generated speech is delivered as 24 kHz, 16-bit mono WAV. That's ideal for voice in video and podcasts. Your editor can convert it to any other format.
Ready to record? Follow the voice sample recording guide.