Make a finished song without any AI service: fix your vocal, get music made for your voice, sing over your own track or karaoke, and master it. Everything works offline in your browser.
AI tools: let AI make music for you to sing over, or write and sing a full song from your lyrics. Needs your AI generator (Colab) connected below.
Only needed for AI music, AI songs and voice improvement. It runs free on a Google Colab GPU: open the notebook that came with this app, press Run, then paste the link and access code it shows.
Raw vocal on its own (WAV, MP3 or M4A). Clean, dry recordings work best.
Drag across the waveform to select a part (a cough, an extra word, noise, a gap), then Cut, Silence or Re-record it. Tap once to move the play position. Zoom in to pick a single word: use the zoom buttons, or Ctrl + mouse wheel / pinch on the waveform; scroll sideways with the wheel or the bar below.
To find the mistake, tap the waveform and press Play with the karaoke. To fix one line: drag across it and press Re-record selection only. To sing everything again from a point: tap there and press Re-record from the cursor to the end, then press Stop when done. Before you sing, the song plays from up to 30 seconds earlier (your old vocal + the karaoke, change it above) so you catch the rhythm, then a countdown tells you when to come in. Only the chosen part is replaced, with smooth joins, at the same place in time. Every fix can be undone.
Plug in or pair a microphone, press Record and sing. Your browser will ask permission once. Use headphones if music is playing, otherwise the music leaks into the recording.
Do this once with the headphones and microphone you sing with: the app plays clicks, you clap (or say "ta") exactly with each click, and it measures the delay of your setup. Bluetooth headphones (AirPods) have a big delay, so measure again if you switch.
After every recording the timing is also checked automatically against the music and fixed. If it still sounds early or late, move the Sync fix: it is applied to your last take straight away (no need to sing again). Higher moves your voice earlier.
Lyrics stay pinned at the top while you record and wait for the intro to finish before moving. Too fast or too slow? Move the speed. While singing, tap the line you are on (or use the ↑ ↓ arrow keys) and the lyrics re-sync from there.
Keeps your melody and words, but re-sings them in a cleaner, steadier version of your own voice. Add a clean 10 to 30 second recording of you, speaking or singing, with no music. Use only your own voice.
AI creates music for your vocal. Describe the style, then press Create. It takes a few minutes. If you don't like it, create another.
The AI takes the reference's instruments, groove and mood and makes new music in your vocal's key and tempo. A song with a singer works too: the AI is asked for an instrumental. 40-60% = inspired by it, 70%+ = very close to it.
Best results: make the music first (no vocal needed: set the tempo and key below, or leave them to the AI), then press "Sing over this music now" and record yourself singing to it in headphones. Real songs are made this way, so your timing and notes follow the music naturally. If you already have a vocal loaded, the music is made in your key and tempo instead.
Move the slider or tap a preset: if a karaoke is already made it is redone with the new setting. "Remove 100%" takes out everything that sits in the middle of the stereo picture in the voice range. A voice that is wide, doubled or full of reverb can leave a faint trace; that is the limit of any browser-only method.
Upload a full song, press the button, and your vocal replaces the original singer. Works best on songs where the singer is in the centre, which is most studio songs.
Tip: mark sections with [verse], [chorus] and [bridge]; if you don't, the app adds them (a repeated stanza becomes the chorus). The language is detected from the script and the singer is told to use that accent. Every press gives a different take: if one is weak, press again rather than changing everything. Hindi works but is weaker than English in this model, so keep lines short and clear.
Optional. With a reference, the AI starts from that song's character (tempo feel, instruments, mood) and writes your lyrics into it. 40-60% keeps your song original; 70%+ stays very close.
Any finished song (WAV, MP3 or M4A). You can change its tempo and pitch, then master it.
Press Start, listen to your vocal and tap along on every beat (the button or the Space key), like clapping to a song. 8 or more taps is enough. The music is then built on your taps, so it is always in time with you.
The music follows your singing beat by beat, so it stays in time even if your tempo drifts. If it feels twice or half as fast as you sang, press Half or Double speed. If the beat lands in the wrong place in the bar, change the start beat.
Detected automatically. If the music feels too happy or too sad, switch Major/Minor. If the beat feels off, adjust the tempo or the timing nudge. For your own karaoke, leave "fit" off if you sang along to it.
Only used when "Music follows my singing" is off: the music keeps a fixed tempo and your vocal is bent to fit it.