Transcribe lyrics from audio with timestamps and song sections

Unveil's Lyric Transcription tool listens to an uploaded track and returns timestamped lyric lines grouped into song sections such as verse, chorus and bridge. It also reports key, tempo and time signature. It needs a free Unveil account, and signed-out visitors are sent to sign-up.

How lyric transcription works

Transcribing lyrics is a form of speech recognition, but singing is harder than speech. Vocals stretch vowels, change pitch constantly, overlap with instruments and are often processed with reverb, delay or harmonies. A model has to separate the voice from the rest of the music, decide which sounds are words, and place each line at the right moment in time.

Unveil sends your audio to Google's Gemini model with instructions to identify every structural section, transcribe each vocal line with a start timestamp, return an empty list for sections with no vocals, and avoid inventing words it cannot hear. The result is returned as structured data and displayed alongside a player.

What affects accuracy

• Mix density: a clear lead vocal over a sparse arrangement is transcribed far better than a vocal buried in loud guitars or synths.
• Delivery: fast rapping, heavy slurring, growls and screams reduce clarity.
• Effects: heavy reverb, distortion, vocoders and stacked harmonies can hide consonants.
• Language and accent: unusual accents, slang and made-up words are more likely to be misheard.
• File quality: low-bitrate or clipped files lose the detail that helps tell words apart.

Preparing audio for the best result

If you have a version with the vocal louder, such as a rough mix or a vocal-forward bounce, use that. A clean lead vocal without effects is ideal. If you only need one section, trim the file so the model is not working across a long instrumental.

Use a full-quality file where possible. The tool accepts audio files in general, so a WAV or high-bitrate MP3 is a sensible choice. Avoid uploading a heavily compressed copy that was already converted several times.

Using the tool

1. Create a free Unveil account or sign in. Signed-out visitors are sent to sign-up.
2. Open Lyric Transcription and drop in an audio file or click to browse. The tool charges 2 ProCoins for a run.
3. Watch the progress while the file is read, uploaded to the AI engine and transcribed.
4. Review the result: sections are shown in order with their start and end times and the lyric lines inside each one.
5. Press play and click a section or a lyric line to jump the audio to that point, then check the text against what you hear.

Reading and correcting the output

Treat the transcript as a first draft. Listen through once with the lyrics on screen and correct anything wrong in your own document. Look especially at repeated lines such as choruses: if the same chorus appears with different wording each time, check which version is actually sung.

Section labels are the model's judgment of the song structure. A pre-chorus or bridge may be labelled differently than you would label it, and the boundaries are approximate timestamps rather than exact bar lines. If you need exact alignment for subtitles or a lyric video, plan to adjust the timestamps manually.

Limits and common mistakes

The most common mistake is publishing machine-written lyrics without proofreading. Misheard words can change a line's meaning, and a missing line can break the rhyme scheme. Another is assuming a section marked as having no vocals really has none, since quiet backing vocals may be missed.

Also remember that lyrics are copyrighted. Use a transcription of your own songs, or songs you have permission to work with, and do not republish lyrics you do not own. Finally, the tool returns BPM, key and time signature estimates from the same AI pass, so verify those in your DAW before relying on them.

What you get

• Lyrics transcribed line by line, each with a start timestamp in seconds.
• Song structure split into labelled sections such as intro, verse, pre-chorus, chorus, bridge, instrumental and outro.
• Track metadata: duration, BPM, key and time signature, plus track and artist name when they can be identified.
• A built-in player with a waveform where clicking a section or a lyric line jumps the audio to that point.

Questions

Related

• Free online LUFS checker
• Free online true peak meter
• Online AI mastering with an analysis-driven mid/side chain
• Mix analysis guides
• Dynamic range analysis: DR, crest factor & LRA