How to Prepare a Clean Voice Sample
VoiceLuma AI Editorial Team · 9/26/2026
The quality of a cloned voice is decided before any AI sees the audio. A clean reference recording produces a better, safer voice model. Here is how to prepare one.
Record in a quiet space
Choose a small, soft-furnished room. Close windows, silence fans and phones, and avoid rooms with hard echo. Background noise that seems quiet to your ear can dominate a model's attention.
Use a consistent setup
Keep the same microphone, the same distance from your mouth and the same volume throughout the recording. Consistency gives the model one clear picture of your voice instead of several.
Speak naturally, not performatively
Read passages that sound like how you actually speak. A model built from an exaggerated performance reproduces that exaggeration in every sentence.
Give it enough material
Read a continuous passage — a few minutes of natural speech covering varied sentences. Reading the same phrase repeatedly teaches the model far less than natural variety.
Do not use someone else's audio
Only submit recordings you made yourself or are licensed to use. Clips lifted from interviews, streams, podcasts or calls do not qualify, and cloning another person without documented permission is prohibited.
Checklist before upload
No background music, no echo, no clipping, consistent distance, natural delivery, and a consent statement recorded if the voice is not your own.
Start a consent-based voice request on the VoiceLuma AI voice cloning page.