How long should a voice sample be for an AI voice cover?
Aim for 10–30 seconds of your voice alone, recorded in a quiet room. Floniks AI Cover accepts samples from 8 to 60 seconds in MP3, WAV, M4A or WebM, up to 20 MB. What you record depends on the mode: to cover an existing song, sing a cappella (any song, no backing music); to make a new song in your voice, just talk naturally. Avoid echo, background music, other voices and voice effects, because a short clean take beats a long noisy one.
The numbers: 10–30 seconds recommended, 8–60 accepted
On Floniks AI Cover the recommended voice sample is 10–30 seconds. The page rejects anything shorter than 8 seconds or longer than 60 seconds, and files must be MP3, WAV, M4A or WebM up to 20 MB. More length is not automatically better: past about half a minute you mostly add more chances for noise, coughs or a second voice to slip in. For "Cover a song", the song you upload has its own limits: at least 10 seconds, at most 6 minutes, and up to 23 MB. If a full track is too large, exporting it as MP3 usually brings it under the limit. You can upload a file or use "Record now" to capture the sample directly in your browser.
Sing for covers, talk for new songs
The two modes want different samples. For "Cover a song", the current cover model uses a sung reference: record yourself singing a cappella, any song you know, with no backing track. A sung sample carries your singing timbre and range, which the model uses to rebuild the vocal and pick the right octave; a spoken sample tends to sound more like reading aloud once converted. For "Sing a new song", the voice is cloned for a song that does not exist yet, so the page asks you to just talk normally for 10–30 seconds. If you save voices for reuse, it can help to keep one sung take and one spoken take and name them clearly.
How to record a clean sample
Record in the quietest, least echoey room you have; a small room with soft furnishings beats a bathroom or a bare kitchen. Keep the phone or microphone close and steady, and use one voice only, with no music playing, no TV in the background and no reverb, auto-tune or other voice effects. The in-browser recorder deliberately turns off browser noise suppression, echo cancellation and automatic gain, because those features clip held notes and breathy tones; the trade-off is that room noise is captured as-is, so a quiet space matters. Sing or speak at a comfortable, natural volume rather than whispering or shouting, and listen back before you submit.
If the cover does not sound like you
The voice sample is the first thing to fix. Re-record a cleaner take before changing anything else: remove background music, move closer to the mic, and sing in a comfortable register rather than straining for the original singer's notes. For covers, the Key setting matters too. "Auto" matches the song to your range and shifts by an octave only when needed, while "Lower octave" and "Higher octave" force the shift when the original singer's range is far from yours. Finally, start from a source track with a clear lead vocal: the cover pipeline separates the vocals from the instrumental before converting them, and a muddy mix separates less cleanly.
Related questions
Build it on Floniks
Image, video, digital humans, and reusable workflows on one canvas. No card required.
Explore Floniks