Karaoke Recorder

About the Karaoke Recorder

Sing over your own backing track and record yourself, all in the browser. Open a song or a karaoke (instrumental) version of it, paste the lyrics — plain text, or an LRC file with a time on every line — and they scroll as you sing, the words lighting up as they come. Record with headphones; a latency test lines your voice up with the music. Then mix the take with reverb, echo and a gentle compressor, punch in a line you want to sing again, and make the mix as MP3 or WAV, or a lyric video as MP4.

Before you sing, the track’s key can go up or down by as much as six semitones and its tempo can run from 75 % to 125 %, so a song fits your voice and your pace; the lyrics follow the tempo. Plain lyrics can be timed by tapping as the song plays, and saved as an LRC file for any music player.

Without a pass, the recorder works in full as a free preview: you record, listen back and punch in as often as you like. The mix played live, and a track whose key or tempo you changed, carry a short MySmartCoPilot chime every 10 seconds; Make the mix plays the first half of the mix (up to 30 seconds) with the chime, and Make the video shows the first half of the video (up to 10 seconds) with a watermark. A pass removes them and unlocks every mix and video download; unlocking one take or one video instead does that for that result alone. Your voice alone, your lyrics and your LRC file are never locked.

Common uses

• Recording a cover of a song over its karaoke (instrumental) version.
• Practising a song for a performance and hearing how your voice sits with the music.
• Making a sing-along lyric video of a backing track for a class, a choir or a party.
• Singing a song lower or slower than the original while you learn it.
• Timing a song’s lyrics into an LRC file for a music player.

Lyrics, LRC files and timing

An LRC file is plain text with a time tag before each line: [mm:ss.xx] is minutes, seconds and hundredths of a second, so [01:02.50] is 62.5 seconds into the song. A line can carry several tags when it repeats (a chorus), ID tags such as [ti:] (title) and [ar:] (artist) describe the song, and [offset:] moves every line by that many milliseconds, a plus sign making the lyrics appear sooner. Enhanced LRC adds word tags such as <01:02.90> inside a line; with them each word lights up at its own time, otherwise the words of a line light up evenly until the next line begins (at most 8 seconds, so an instrumental gap does not crawl). In an LRC file, lines without a time are left out, and the page says how many.

Lyrics without times are timed by tapping: start timing, and press Next line — or Enter — just as each line begins; Back a line undoes a tap. The times go into the lyrics as LRC, which Copy as LRC and Download the LRC save. If every line comes out a little early or late, Move every line shifts them all by 0.1 or 0.5 seconds.

Why a recorded voice comes out late, and the latency tests

Your voice reaches a recording later than the music by two delays: the music’s way out to your ears (what the browser reports as its base and output latency) and your voice’s way in through the microphone (the input latency the microphone track reports). You sing in time with what you hear, so without a correction the voice sits late in the mix, often by tens of milliseconds and by far more with Bluetooth.

The click test measures both delays together: eight clicks play at uneven gaps (so an echo is never taken for the next click) while the microphone listens, the start of each click is found in the recording, and the result is the median delay, accepted when at least six clicks lie within 12 ms of it. The clap test has you clap with ten steady clicks, after the first two; it needs six claps within 45 ms of their median and is less exact, as no one claps exactly on the beat. Without a test, the page uses the delays the browser reports. The mix moves your voice earlier by the result, plus your own adjustment of up to 300 ms either way. Bluetooth headphones and speakers delay the sound by much more than wired ones, and the browser’s own figure for its output can change while the page runs — wired headphones give the tightest takes.

Key and tempo

The key and tempo are changed with Signalsmith Stretch (MIT licence) running in your browser: the key up or down by up to six semitones (one semitone is one piano key) without changing the speed, and the tempo from 75 % to 125 % without changing the pitch. The whole track is rendered before you sing, which takes a little while for a long song, and the lyrics’ times follow the new tempo. Takes are sung to the track as it is, so changing it again removes them (the page asks first). Small changes sound natural; large ones sound processed.

The mix

The mix plays your take with the track: Voice and Track at 100 % are as recorded and as the file is, up to 150 %. Reverb sends the voice into a generated room of about two seconds; Echo repeats it at the echo time, each repeat about 8 dB softer and darker; the compressor turns the voice down gently (3:1 above −24 dBFS, with a soft knee) so loud and soft lines sit closer together; and a limiter stops the mix from clipping. What you hear with Play with the track is exactly what the file holds: the same audio graph renders both.

A mix covers the part of the song the take covers, plus 2.5 seconds of the reverb and echo ringing out, as a 192 kbps MP3 or a 16-bit WAV at the sample rate the song plays at on this device (shown beside the song once it is open). Download my voice saves your voice alone, lined up the same way and starting where the mix starts, to mix in another program. Up to three takes stay on the page (the oldest goes first); a four-minute take holds about 45 MB of memory.

Lyric videos

A lyric video shows the title and artist, the line being sung with its words lighting up, the line before it and the two after, on a colour wash or your own picture under a dark veil. Choose 1280 × 720 (16:9), 720 × 1280 (9:16, upright, for phones) or 1080 × 1080; the video runs at 30 frames a second, with the track alone (a sing-along video) or a take mixed with it. It is encoded in your browser with WebCodecs: MP4 with H.264 video and AAC sound where the browser can, otherwise WebM. Making a video needs the browser’s WebCodecs video encoder, which current Chrome and Edge have; the page says so if this browser has none. Arabic and Hebrew lines light up from right to left.

Sources

• Wikipedia: LRC (file format)) — time tags, ID tags, the offset tag and word tags
• W3C Web Audio API: baseLatency and outputLatency — the delay out to the speakers
• W3C Media Capture and Streams: latency and echo cancellation — the microphone’s delay and processing
• Signalsmith Stretch — the key and tempo changer (MIT licence)

Limitations

• Use songs you have the right to use; sharing a cover or a lyric video publicly can need permission from the song’s rights holders.
• The track plays as it is: use an instrumental (karaoke) version, since with the original recording its singer stays in the mix.
• With speakers instead of headphones the microphone records the music too; echo cancellation reduces it but colours the voice.
• Bluetooth headphones and speakers add a long delay that can change while you sing; run the click test with them, or use wired headphones.
• Songs of up to 12 minutes. Takes live in this page’s memory only (at most three): leaving or reloading the page loses them, so save what you want to keep.
• Changing the key or tempo of a long song takes a while, and big changes sound processed.
• Lyric videos need the browser’s WebCodecs video encoder, which older browsers lack.
• Phones may stop recording when the screen locks or you switch apps.

Privacy

On this device; its files download once

Everything happens in your browser. What you enter or open here is not uploaded or stored by MySmartCoPilot. Parts of the tool (the decoder, the key and tempo changer, the MP3 encoder and the video’s font) download from MySmartCoPilot the first time you use them. Your songs, recordings and lyrics never leave this device.

Open-source components

This page can load these open-source components, each only when a task needs it:

• It uses FFmpeg under the LGPLv2.1 and the LAME MP3 encoder under the LGPL, through libav.js: licence and source code.
• It uses libheif and libde265 under the LGPL v3, through heic-to: licence and source code.
• It uses FFmpeg under the LGPLv2.1, through the Mediabunny add-ons: licence and source code.
• It uses the LAME MP3 encoder under the LGPL, through the Mediabunny MP3 add-on: licence and source code.
• It uses Mediabunny under the MPL 2.0: licence and source code.
• It uses the TurboRes ProRes decoder under the MPL 2.0: licence and source code.

You may replace them with your own builds. Their licences and the credits of every library the site uses are on the open-source credits page.