AI vocals off beat? How to fix it

An AI vocal usually sounds off beat for one of five reasons: the tempo was read at half or double the real BPM, the beat file was messy, the lyrics did not fit the bars, the vocal starts in the wrong place, or the mix hides the timing. Fix them in that order. Check the BPM number before you generate, give the tool a clean WAV, write lines with a steady syllable count, slide the vocal so its first word lands on the first downbeat, then bring it up a little, add a touch of reverb and master to -14 LUFS. Plan on two or three tries for a take you keep. In Lyro the vocal comes back as its own track at your beat’s tempo and key, and you line it up on the studio’s snap grid.

Drop your beat to read its tempoWAV, MP3, FLAC, M4A or OGG, up to 60 MB. Tempo and key are read in your browser; nothing is uploaded until you sign up.

No file at hand? Start from a description, or generate a prompt.

How it goesThree steps

Check the tempo number firstLyro reads your beat’s BPM in the browser. If it shows half or double the real tempo, such as 70 for a 140 BPM trap beat, type the right number before you generate. A wrong number makes the vocal drag or rush.
Give it a clean file and lyrics that fitUpload a WAV that starts on the first bar and keeps one steady tempo. Write a full verse and a chorus, with lines of a similar length, so the singer never has to cram or stretch.
Line it up, mix it, try againIn the studio, slide the vocal until its first word sits on the downbeat and trim the silence before it. Turn it up a little, add a touch of reverb and master to -14 LUFS. Expect two or three takes.

The details

Why AI vocals land off beat

When Lyro sings your lyrics over your beat, the beat itself is not sent to the music model. Lyro reads its tempo and key in your browser, asks the model for a sung performance at exactly that tempo and key, then keeps only the voice and returns it as an isolated vocal track. Your beat stays untouched, and that has two consequences for timing.

The vocal can only be as right as the tempo number. If the number is wrong, the vocal is phrased at the wrong speed.
The model does not know where your beat’s first bar starts. The vocal comes back in tempo, and you decide where it begins, in the studio.

The other causes, lyrics that do not fit the bars and a mix that blurs the timing, apply to any AI vocal made with any tool.

1. Check the tempo number: half or double

Tempo detection listens for the pulse of the drums, and it can lock onto a pulse exactly half or double the real one. A trap beat made at 140 BPM and felt in half-time can be read as 70; a slow groove with busy hi-hats can be read at twice its speed. The number looks plausible, which is why it is easy to miss, and the vocal then drags, rushes or only meets the beat every other bar.

If you made the beat, use the tempo from your project. It is exact.
A leased beat often lists its BPM on the store page or in the file name.
Compare the reading with the usual range for the style: trap is made at about 130 to 160 BPM and felt at half that, boom bap sits around 85 to 95, house around 120 to 128. A reading at half or double those is suspect.
Check the key while you are there. It can land on the relative key, such as A minor for C major. That does not move the timing, but it can make the vocal sound wrong. See AI vocal out of key.

You can read any file’s tempo and key with the BPM and key finder, in your browser. A vocal that sounds twice as fast or half as fast as your beat almost always means this number was wrong, and mixing will not fix it: correct the tempo and generate again.

2. Give it a clean file

Export a WAV from your music software that starts on the first bar, not on a count-in or a long stretch of silence.
Keep one steady tempo. A beat that speeds up, slows down or changes tempo halfway cannot be described by one number, so the vocal will drift somewhere.
Leave some headroom. A beat pushed into a limiter until it distorts is harder to read.
If you have a version without a producer tag or a guide vocal on it, use that. A voice in the file confuses the tempo and key reading.

A new account can try it once for $1, without a subscription, on the first 60 seconds of its own file. Put the part you want sung over at the start of the file.

3. Write lyrics that fit the bars

A vocal falls behind when a line holds more syllables than the bar has room for, and it floats when a line is too short. The model then crams, stretches or rearranges your words to make them fit.

Write a full verse and a chorus, marked with [Verse] and [Chorus]. With only four lines or so, the model adds lines of its own to fill the song. If you want to hear only your words, give it enough of them.
Keep the lines of a section to a similar length. For sung lines, around six to ten syllables is comfortable at most tempos; rap bars can hold more.
Count them with the syllable counter, which shows syllables per line and the rhymes at line ends (the count is an estimate for English).
Keep the chorus word for word the same each time it returns.
Read the lyrics aloud over the beat. If you trip over a line, the model will too.

With no lyrics at all, Lyro improvises a topline instead: a melody that follows the beat, not your words. See the AI topline generator.

4. Put the first word on the first downbeat

The vocal comes back as its own track at your beat’s tempo. Lyro does not know where your beat’s first bar starts: an intro, a pickup or a moment of silence at the top of the file all move it. So the last piece of timing is yours, and in the studio it takes a minute.

Keep Snap on, so clips land on the beat grid.
Trim the silence at the start of the vocal clip by dragging its left edge to just before the first word.
Drag the vocal until that first word sits on the downbeat of the bar where the verse should start. Play from the bar before it and listen.
If one section still sits early or late, split the clip there and move that part on its own.
If the whole vocal slowly slides away, the tempo is slightly off. The vocal tools can fit a clip to the project tempo when the difference is small; for half or double, go back to step 1.

5. Mix it so the timing reads

A vocal that is buried or drowned in reverb sounds loose even when it is on time, because the start of each word gets lost.

Bring the vocal up a little, until every word is easy to follow over the drums.
Add a touch of reverb, not a wash. A little space sits the voice in the beat; a lot blurs the timing.
Master to about -14 LUFS with a ceiling near -1 dB, the streaming preset in the studio, and compare before and after with the A/B switch.

Mixing and mastering run in your browser and use no credits. More in mixing and mastering.

6. Plan two or three tries

The same lyrics on the same beat give a different performance every time, and one take often sits better than another. Change one thing between tries, so you can hear what it did, and budget for two or three. On Lyro High a vocal over a beat of up to a minute costs 7 credits, and 11 credits over a three-minute beat; an improvised topline costs one credit per started minute, with a minimum of 2 credits. Starter includes 20 credits a month for $9, which covers a vocal over a one-minute beat twice. If a generation fails, its credits come back automatically.

When the beat drifts under your vocal

The same checks apply the other way round, when Lyro writes a beat around your vocal, with one difference: a first try can drift in tempo against your voice. Check the tempo reading, then run it again. Singing to a click in headphones gives the music a steadier pulse to follow. A new run costs one credit per started minute, with a minimum of 2 credits. See add a beat to your vocals and turn a voice memo into a song.

Watch it done

Our step-by-step video, How to add AI vocals to your beat (or a beat to your vocals), walks through both directions in Lyro, step by step, on YouTube.

Questions

Why are my AI vocals off beat?

Most often the tempo number: detection can read half or double the real BPM, and the vocal is then phrased at the wrong speed. After that come a messy beat file, lyrics with too many or too few syllables for the bars, and a vocal that starts in the wrong place. In Lyro the last one is quick to fix: slide the vocal in the studio until its first word lands on the first downbeat.

Why does the vocal sound twice as fast, or half as fast, as my beat?

It was generated at half or double your beat’s real tempo. Detection sometimes locks onto half or double the pulse, especially on trap and drill beats with a half-time feel. Type the real BPM before you generate and run it again. The studio’s time-stretch is made for small differences, not for a factor of two.

Why did the AI add lines I did not write?

With only a few lines, around four, the model fills the rest of the song with lines of its own. To hear only your words, write a full verse and a chorus with [Verse] and [Chorus] tags, and keep the chorus identical each time it returns. The syllable counter helps you keep the lines even.

Does Lyro know where my beat’s first bar starts?

No. It reads the tempo and key, not the arrangement, so an intro, a pickup or silence at the start of your file is invisible to it. The vocal comes back as its own track, and you place it in the studio: keep Snap on, trim the silence before the first word and drag the vocal until that word sits on the downbeat.

Can Lyro fix the timing of a vocal I recorded myself?

Partly. In the studio you can trim, split and move clips on the beat grid, and fit a clip to the project tempo when the difference is small. Lyro does not re-time a performance word by word. If you would rather have music that follows your own timing, Lyro can write a beat around your vocal instead.

How many tries does it take?

Plan on two or three. Each generation gives a different performance. On Lyro High a vocal over a beat of up to a minute costs 7 credits, and credits for a failed generation come back automatically. Starter is $9 a month for 20 credits. See the pricing page.

Hear it for yourself.

Try it Generate a prompt