Recording tips for better AI voice changer results

By the voicechange.io team · November 6, 2026

An AI voice changer can only work with what it hears. Almost every "it sounds robotic" or "it sounds like it's underwater" result comes from the input recording rather than the voice model. Here is what matters, in order of how much difference it makes.

Why the input matters so much

Voice conversion models like the ones on voicechange.io (RVC models) don't understand words. They look at your recording a few hundredths of a second at a time, work out what sound you are making and at what pitch, and redraw that sound in the target voice. Anything that is not your voice, such as a fan, music, an echo or a second person, gets treated as if it were part of your voice and is redrawn too. That is where the metallic warbles and garbled consonants come from.

1. Remove background noise and music

This is the single biggest factor.

Record in the quietest room you have. Turn off fans, air conditioning and the TV.
Never record over music. If you want to sing along to a track, use the Song cover mode with the original song instead; it separates the vocals properly.
Avoid rooms with hard, bare walls. A bedroom with a bed, curtains and a carpet sounds far better than a kitchen or bathroom, because echo blurs the edges of every syllable.
Phone noise suppression (on by default in most voice memo apps) is fine. Heavy noise removal plugins applied afterwards often make things worse, because they leave artifacts the model then copies.

2. Get the distance to the mic right

Hold your phone or microphone about a hand's width (10 to 20 cm) from your mouth, slightly to the side so plosives ("p" and "b") don't pop into it. Too far away and the room echo becomes louder than your voice. Too close and the recording distorts on loud words.

3. Watch your levels

If your recording is very quiet, the model struggles to hear the pitch and you get a wobbly, breathy result. If it is too loud and clips (the waveform hits the top and flattens), the distortion gets converted into crackles. Speak at a normal, confident volume. If you are using recording software, aim for peaks around three quarters of the meter.

4. Speak clearly and naturally

Act it out. The model keeps your timing, emphasis and emotion. If you want an excited Pomni or a menacing Darth Vader, perform it that way; a flat read gives a flat result.
Don't whisper. Whispering has no pitch, so the model has nothing to convert and produces noise. Soft speaking is fine.
One speaker at a time. Two voices at once will be blended into one strange voice. Record each person separately.
Leave a short pause at the start and end so the first and last syllables are not cut off.

5. Let auto pitch do its job

If you have a deep voice and pick a high character voice (or the other way round), the pitch has to move. Auto pitch measures your recording and shifts it toward the voice's natural range. Turning it off and forcing your own pitch through usually makes high voices sound strained and low voices sound muffled. If the result is close but slightly off, nudge the slider by one or two steps. The pitch guide explains when to move it further.

6. Choose formats and lengths sensibly

WAV gives the cleanest input, but an mp3 at 192 kbps or more is indistinguishable for this purpose and much smaller. Uploads are limited to 15 MB.
Short clips (10 to 60 seconds) are faster to iterate on. Get one sentence sounding right before converting a long take.

Quick troubleshooting table

What you hear	Most likely cause	Fix
Metallic, robotic warble	Background noise or music	Re-record in a quiet room, no music
Voice cracks or jumps an octave	Pitch too far from the voice's range	Turn auto pitch on, or move the slider 12 steps
Muffled, "underwater" sound	Pitch shifted too far down, or a lot of room echo	Raise the pitch a few steps; record in a softer room
Crackles on loud words	Clipping in the input	Record a little quieter or further from the mic
Breathy, unstable result	Recording too quiet, or whispering	Speak up and move closer to the mic
Two voices merged together	More than one speaker	Record each speaker separately

Still not right?

Some voices are simply better models than others, because they were trained on more or cleaner audio. Each voice page has a sample clip made from the same sentence, so you can compare them before converting. If a particular voice consistently sounds wrong, let us know.

Try it now: the voice changer is free and needs no account. Pick from all voices, or read the FAQ.