WaveMix·Your AI soundtrack editor

On most tracks you can’t hear the edit. It avoids cutting where someone is singing and matches the level across the join, so it just sounds like the track was always that long. Music with a vague pulse and no clear drums is where it slips. If you don’t like the result there are three ways out: take a different option (you always get several, and you can audition them side by side), drag that one cut somewhere else yourself, or change the target length by a few seconds and run it again — a few seconds is often all it takes to find a better landing spot.

Yes — there are only three steps: drop in a track, type the length you want, and audition the options. It works out the rest. The tempo and key readouts are there for people who want to check them; ignoring them changes nothing about the result.

That depends on the licence for the track you brought, and has nothing to do with this site — we only change its length and its joins, and we do not grant any music licence. There is no music library here either; the track is one you uploaded. Rights stay with the rights holder, and commercial use is yours to clear. One more thing worth saying: changing a track’s length does not make a copyrighted song free to use.

No. The audio stays on your own computer from beginning to end, and the analysis runs there too. Once you sign in, the only thing that goes to the cloud is the analysis result and your edit — a small text record with no audio in it. So when you open it on another computer everything is waiting, but you have to pick the audio file again.

On most tracks, yes. When it doesn’t, just fix it: drag on the waveform to move where a section starts or ends, click the label to change “chorus” to “interlude”. Everything generated afterwards follows your version — nothing needs re-analysing, only the part you touched.

It depends on the length of the track and on your computer — usually a few minutes, longer on an older machine. You are not staring at a bar: results fill in one at a time, tempo and beats first, then the section map. Once a track has been analysed the result stays on your machine, so opening it again is instant.

MP3, WAV, FLAC, AAC and M4A, up to 50 MB per file. 50 MB is roughly a ten-minute MP3; as WAV you’ll hit the ceiling at three or four minutes, since WAV is uncompressed and much larger for the same length. If you go over, convert to MP3 first.

Both. It transcribes English and Chinese lyrics, and it is more accurate when the singing is clear and the backing isn’t too loud. Anything it mishears can be proofread by the AI, or you can paste the official lyrics in and replace the lot at once. Instrumentals have no lyrics, so sections, beats, retiming and blending all work as usual — you just lose the layer that keeps cuts out of the middle of a sung phrase.

Yes. The AI runs on your own model account: fill in three things in settings — endpoint, model name and key — and it works. The key is stored only in your browser and never uploaded. Whatever it costs is between you and your model provider; we neither handle nor mark up the charge. Leaving it blank costs you nothing else: sections, beats, lyrics, retiming and blending all work as usual, only without the AI’s suggestions.

Lyrics export as an SRT file, which every major editor imports as subtitles — drop it on the timeline. Section positions export as a marked WAV or a CUE sheet. One thing has to line up: set the frame rate in settings to match your editing project before exporting, or every marker lands offset. Whether the markers can be read directly also depends on which editor you use.

Chrome and Safari, on a reasonably recent version. Other browsers may well open it, but they haven’t been tested one by one. It runs on phones and the layout adapts, but the analysis is demanding, so a phone is a good deal slower and dragging around a waveform is fiddlier than on a desktop — for real work, use a computer.