Enhanced LRC Maker: Word-Level Karaoke Lyrics from Your Audio

Most LRC makers work the same way. You paste the lyrics, press play and hit a key when each line starts, and the app writes the timestamp. For lines that's fine. A song has maybe 40 to 80 of them and you get a second or two of warning before each one.

Words are harder. A 3 minute pop song has 300 to 500 words. In a fast verse they come 4 or 5 a second. Your reaction time is around 200 ms on a good day, so by the time you've tapped one word the next two are gone. People do try. They slow playback to half speed, tap, then double the timestamps. That takes an hour per song and still lands late.

So tap-to-sync editors stop at the line level. In most of them the "enhanced" option is a text box where you type the angle-bracket times yourself, which is the hour-per-song method again with a nicer font.

A speech model doesn't react to the vocal. It reads the whole waveform and works out where each word starts and stops. The word boundaries come out of the same process that produces the transcript.

Roughly it goes like this. The audio gets split into short frames and the model labels each frame as part of a word or as silence. Runs of frames with the same label become words with a start and end time. The line timestamp is the first word's start and the angle bracket tags are each word's start. Nobody taps anything.

Accuracy is tightest on clear lead vocals. Sung words stretch, so where a long held note "starts" is a judgment call and you might pull it a touch earlier than the model did. Backing vocals and ad-libs sometimes get stamped as their own words. Both take seconds to fix on a waveform, which beats typing 400 timestamps.

Support is narrower than for plain LRC. Here's what we could find in each project's docs, forum or on Wikipedia.

If your player isn't listed, test it. Drop an enhanced file in and watch one line. Words lighting up one by one means it works. If the whole line highlights at once the player ignored the tags, and that's harmless. But if you see <00:12.71> printed as text the player choked on them, so give it plain LRC.

The enhanced LRC word-level timing guide goes into each one.

Enhanced LRC is the simplest word-level format. It isn't the only one though. If you're targeting a video or a web player you might want one of these. Both can carry word timing.

ASS. The subtitle format Aegisub and most anime fansubs use. Karaoke is built in. Each word gets a {\k} tag with its duration in centiseconds and the renderer fills the text from left to right.

Dialogue: 0,0:00:12.40,0:00:15.10,Default,,0,0,0,,{\k31}I {\k24}left {\k25}the {\k42}porch {\k36}light {\k32}on {\k25}for {\k55}you

Use this for a lyric video or anything that plays through a video player with libass. The free LRC to ASS tool converts line timing only, so the {\k} tags are something you add in Aegisub or with a karaoke script.

TTML is XML timed text. Each word becomes a <span> with its own begin time inside a <p> for the line. Produce this if a distributor or a web player asks for timed text and not a lyrics file. The free LRC to TTML tool writes one <p> per line and drops the word tags. Word-level TTML needs a tool that writes spans.

EasyLRC exports all of these formats from one timing pass, so there's only one review. Right now the ASS and TTML exports are line-level. The word timing goes out as Enhanced LRC.

• Enhanced LRC word-level timing guide
• Enhanced LRC format examples
• LRC format complete guide
• Best LRC editors in 2026
• Fix lyrics time drift in LRC files