Songs
No song yet. Add an MP3 or WAV to sing over, or drop one anywhere on the page. Add more and they stagger on the mix's timeline.
Selected song
- From
- 0:00.0
- To
- 0:00.0
Each song is a track of its own: a level, Mute and Delete on its row, and a place on the mix's timeline (a new song lands at the playhead; drag it in Mix to move it). Select one to edit it. Editing a song never moves the takes: only a drag, a nudge or an offset of your own moves anything on the mix. A take remembers where the playhead was when it began; the songs play along only if asked for, in Record. A take's own Play plays the take alone; Play mix plays everything together.
Record
Headphones on: the count-in and playback come out of the speakers otherwise, straight into the microphone. R records, space plays, [ and ] set the trim at the playhead.
Your voice
Read forty short lines, one at a time, and the booth bundles them for a voice that sounds like you. The lines are made up for this; read each plainly, as you would say it, in a quiet room with the microphone close.
While the script is open, Record (and R) records the line shown, with no count-in and no song; each reading is a take named by its line, and the newest reading of a line is the one that counts. The bundle is a zip of the readings as 22 kHz WAVs with a metadata.csv, the layout tools/train-voice/ in the project's source turns into a voice file for the Words card. Nothing leaves this browser until you download it.
Words
The voice is made on this machine: the first use of a voice fetches it (about 60 MB) and keeps it in this browser. Each line becomes a take at the song's playhead with the Tune settings of the moment, so a melody typed in Tune is sung. Nothing you type leaves this browser. Add a voice file puts a voice of your own in the list: the .onnx and .onnx.json that tools/train-voice/ makes from the Your voice card's bundle, kept in this browser only.
Takes
Nothing recorded yet. Open the microphone and press Record, or type a line in Words.
Selected take
- From
- 0:00.0
- To
- 0:00.0
- Length
- 0:00.0
Tune
The robot plays on the way out, live and in the mix; the vocoder's tone follows the note the tuner aims at. Retune at 0 snaps every note hard; towards 400 ms it eases in like a steady singer. Humanize keeps some of the natural drift. Shift moves everything by semitones. A melody replaces the key: each time the voice comes in after a gap it moves to the next note, and the last note holds (or the melody repeats). Settings belong to the selected take and to the next one recorded.
Mix
Add a song or record a take, and they line up here.
The songs on top, a lane each; under them, each take where it sits, after its trim, latency and nudge. Drag a song or a take to move it; click anywhere else to move the playhead. Play mix plays every song and take that is not muted, each tuned as set, and Download mix saves the same. Make a video records the mix as a vertical video for a phone, the site's card on top and the waveform moving with the music, as it plays, silently; the title on the card is the one typed beside it, or the song's name.
Account
No snapshot in your account yet.
A snapshot is the whole booth as it is now, songs, takes and settings, kept in your account so it is not lost with this browser's storage. The sound goes up as 16-bit WAVs, about 10 MB a minute for a stereo song and 6 MB a minute for a take, so a full project is tens of megabytes and the first save takes a minute or more on a home connection. Open puts a snapshot in the booth in place of what is here, and the booth then works on it: Save makes a new version of it, sending only the sound that changed, and Back to last save loads the version opened or saved last again. Earlier versions can be opened too. A snapshot is not kept up to date by itself: save after a change.