Song to lyrics

Get the lyrics of a song from the recording itself: the vocals are separated from the music, then transcribed with timestamps, on your device. Free, nothing uploaded.

AudioRinse · lyrics transcriberOn-device · 0 B uploaded

Drop a song here

MP3 · WAV · M4A · MP4 · MOV · WEBM — up to 3 hours. Nothing is uploaded.

Presets

The vocals are separated from the music first, then transcribed: Accurate hears sung words much better than Fast. The models download once and stay in this browser, so later transcripts start at once and work with Wi-Fi off. Your song is never uploaded. Free: Fast lyrics as text, SRT or VTT, no limit. Pro: Accurate lyrics, which get far more sung words right, and Word files. Your first Pro download is free with an account.

How it works

  1. Drop a song onto the box above: MP3, M4A, WAV, or the sound of a video. Fast is free; Accurate gets more sung words right (Pro).
  2. AudioRinse separates the vocals from the music, then Whisper writes down what's sung, all on your device. The first time, the models download once.
  3. Read along: click a line to hear it in the song, and correct any word in place.
  4. Download the lyrics as text, or as SRT or VTT with each line's time, for a lyric video.

How it hears the words

Speech recognition is trained on talking, and music under a voice throws it: drums and guitars get written down as words, and quiet lines go missing. So AudioRinse first runs the vocal remover's separation model and keeps the vocals alone, then gives those to OpenAI's Whisper. Every line keeps its time in the song.

We measured it on 14 song excerpts from MUSDB18, a standard test set of songs with their vocals recorded apart. Of the 147 words Whisper (Fast) hears in the vocals alone, the separated mix gave back 123, and the same model on the full mix 117.

Both models run in your browser. The song never leaves your device, and after the first time the page works with Wi-Fi off.

What comes out well, and what doesn't

  • Clear lead vocals do best: pop, folk, singer-songwriter, musicals, most rock.
  • Fast rap, growled or screamed vocals, heavy effects, choirs and stacked harmonies are hard, and backing vocals and ad-libs are often left out.
  • Accurate, Whisper large-v3-turbo, is a much larger model than Fast and gets more sung words right; it needs a graphics chip with WebGPU to run at a good speed.
  • Check the result against the song: click a line to hear it, and fix the words in place before you download.

Lyrics for a lyric video or karaoke

Download SRT or VTT and each line comes with its start and end time, ready for a video editor or a player that shows subtitles. For a karaoke track, take the vocals out with the karaoke maker and put the lyrics over the instrumental.

Questions

Is my song uploaded anywhere?

No. The separation model and Whisper run inside your browser, on your own computer. The song and its lyrics never leave it; the only downloads are the models, the first time.

Is it free?

Yes, with the Fast model: lyrics as text, SRT or VTT, with no limit and no account. Pro ($7.99 a month or $39.99 a year) adds the Accurate model, which gets more sung words right, and Word files. The same Pro covers every AudioRinse tool.

Can it find a song's published lyrics by its name?

No: it writes down what it hears in the recording you drop, which works for songs that have no published lyrics, like your own, a demo or a live take. For a well-known song, its published lyrics are a search away, and the transcript is a way to check what's actually sung.

Which languages does it understand?

Whisper transcribes around 100 languages, most accurately English and the major European languages. Pick the language in the menu, or let it detect it from the start of the song.

How long can the song be?

Up to 15 min. On a computer with little memory the page takes less, and says so.

Can I use the lyrics I get?

Song lyrics are usually copyrighted by their writers. Transcribe them for your own listening, study or practice, or for songs you have the rights to.

Other tools