LyricVideoMaker

How to Sync Lyrics to Music (Line-Level vs Word-Level)

How lyric timing works: line vs word sync, doing it by hand or with AI, and the spots where automatic sync usually goes wrong.

LyricVideoMaker

"Sync the lyrics" sounds like one task. In practice it means two very different amounts of work, depending on whether you are timing lines or timing words.

Line-level sync gives each line a start time. A plain LRC lyrics file works this way: the line appears, you read it, the next one replaces it. It is forgiving. If a line shows up a quarter second early, nobody notices.

Word-level sync gives every single word its own start and end time. That is what you need for karaoke-style fills, kinetic text where words pop in on the beat, or a typewriter effect that types along with the singer. It is much less forgiving, because the viewer's eye is following each word, and a word that lands late feels wrong even if they couldn't tell you why.

Decide which one you need before you start, because the methods differ.

Doing it by hand

You can do this in any video editor. I did it this way for years, so here is the honest version.

For line-level sync:

  1. Put the song on the timeline and zoom in on the waveform.
  2. Play the song and tap a marker key at the start of every line. Most editors let you drop markers while playing. Don't stop to be precise. You'll fix them after.
  3. Go back through and drag each marker to the exact start of the first syllable. The waveform helps: you can usually see the vocal transient.
  4. Create a text layer per line that starts at its marker and ends at the next one.

For a three-minute song with 40 lines, expect an hour if you are new to it, maybe 20 minutes once you're quick.

For word-level sync by hand: the same process, but per word. For that same song, you might be looking at 300 or more words. It's tedious enough that most people who try it once go looking for a tool afterwards. I don't recommend it unless the video is short or you need total control for a specific artistic reason.

A tip if you do it by hand anyway: time the lyric to the start of the consonant, not the vowel. "Stay" begins when the "s" begins. If you place the word at the loud vowel, the whole video feels a hair behind.

Doing it with AI

The modern way is to let speech recognition do the first pass. Our lyric video maker works like this:

Karaoke-style lyric video frame made with LyricVideoMaker: the words of "running every red light home" fill with yellow as they are sung, next line waiting below

  1. You upload the song (MP3, WAV or M4A, up to 10 minutes) or paste a public Suno link.
  2. The vocals get separated from the instruments, so the recognizer isn't fighting the drums and guitars.
  3. WhisperX transcribes the isolated vocal and gives every word a start and end time.
  4. If you pasted your own lyrics, the tool aligns your text to the singing instead of trusting what it heard.
  5. Words it's unsure about are flagged, so you check those and leave the rest.

Step 4 is the one I'd underline. Speech recognition on singing is good but it is not as good as it is on speech. Singers stretch vowels, slur consonants, and sit under reverb. When you give the tool the correct words, the hard part of the problem (figuring out what was sung) is gone, and it only has to place known words in time. That is why pasting lyrics makes such a difference to accuracy.

Where automatic sync goes wrong

After a lot of songs run through AI alignment, the failure spots are predictable. Check these first.

Pickups and breaths before a section. A singer takes a big breath before the chorus, or sings a quick "oh" pickup. The aligner can attach the first word of the line to the breath and put it a little early.

Held notes. One word held across a long note. The start time is usually right; the end time sometimes cuts off before the note does. Drag the end out so the word stays on screen for the whole note.

Fast rap. Lots of words, short gaps, consonant clusters. Line placement is usually fine, individual word boundaries less so. If you're using a word-by-word style, give dense verses an extra look.

Repeated lines. If a chorus line appears six times and your pasted lyrics only list it four times, the alignment has to stretch your text over audio it doesn't match. Write out every repeat exactly as it's sung.

Backing vocals and ad-libs. The vocal separation pulls out the backing vocals too. If you don't want "yeah, yeah" on screen, delete those words. If you want them, consider putting them in a smaller line.

Heavy effects. Vocoders, pitched-up hooks, and distortion make everything harder. Expect more flagged words and give those sections a careful listen.

Fixing timing in the editor

In our tool, you can edit both the text and the timing of each word in the browser. My order of operations:

Kinetic lyric video frame: one big coral word, "RED", with the previous and next words small above and below

  1. Go through the flagged words first. Play each one, fix spelling or drag its timing.
  2. Fix the first word of every section. That's where alignment is most likely to drift.
  3. Watch the whole thing once at normal speed, in the style you plan to export. A word that is 100 ms off is invisible in a Minimal line-by-line style and very visible in Karaoke.

The last point is worth knowing. The style changes how much precision you need. Karaoke and Kinetic expose word timing directly. Minimal and Stack mostly show lines, so small word errors hide. If you're short on time, pick the style first and check timing at the precision that style needs.

Beat sync is a separate thing

People mix these up. Lyric sync is about matching words to the voice. Beat sync is about matching animation to the drums: text that pulses, or words that pop on the kick.

In our tool, beat sync is optional. It's great on music with a steady pulse, like dance, hip-hop and most pop. On songs with free tempo (a solo piano ballad, a live recording that speeds up), it can make the text twitch for no reason. Turn it on, watch the chorus, keep it only if it helps.

A quick sanity check before export

Mute the audio and watch the lyrics alone for 20 seconds. Then play with sound. If the text rhythm felt natural even without sound, the timing is good. If it looked jerky, look for lines that are too long or words with tiny durations.

Then play it on your phone. Most people will watch on one. Timing that looked fine on a big monitor can feel rushed when lines wrap into two on a 9:16 screen.

Which method should you use?

If you need line-level sync for one short clip and you already know your editor, do it by hand. It's fine.

For a full song with word-level effects, use AI for the first pass and spend your time fixing the handful of flagged words. That is where your ears are actually useful.

For a full walkthrough from file to finished video, see how to make a lyric video. If the song came out of Suno or Udio, the AI songs guide covers a few things specific to generated vocals.

← All guides