Have an AI sing your lyrics

Choose a voice, paste the words you wrote, and an AI singer sings them in the order you wrote them. The voice is made by AI: it isn’t a recording of a real singer, and it won’t imitate one.

The first song needs no account; each run gives you two versions in about a minute.

Voice
Use words you wrote: lyrics from released songs are refused (copyright).0 / 5,000

More options
Pro fine-tuning
Choose voice gender, style strength, creativity, length, and styles to avoid—Unlock with Pro →

Your songs

Your generated songs will appear here.

Paste your lyrics, pick a style, and press Make my song.

By · Updated · How we write and test

What the AI singer does with your words

It sings what you wrote, in your order, to a melody it makes up in the style you pick. We checked that with 22 lyrics of our own, two takes each. In 38 of the 44 takes every word was sung just as written. Four takes sang the chorus’s last line one extra time at the end, and our longest lyric, 981 words, ran out at about 7 minutes with two lines unsung.

It won’t correct you: a typo or a clumsy line is sung as it stands, so read the lyrics once more before you press the button. And it won’t sing words from a song someone else released, or take a real singer’s name as its style; what you can’t use says what happens if you try.

Four voices, the same eight lines

The buttons above the lyrics box ask for a kind of voice (female, male, duet, choir, rap, soft, whisper), and Auto leaves it to the style. To hear what the choice changes, we gave the same eight lines, a short pop song of ours, to four of them:

“Saturday Morning”, our eight lines

[Verse]

Saturday morning, the kettle’s on half of the city is still asleep I found the note that you left by the phone three little words that I’m going to keep

[Chorus]

Keep it simple, keep it near everything I need is here say it softly, say it twice Saturday morning, you and I

Female
Takes of 1:06 and 0:47, every word sung in both
Male
Takes of 1:00 and 0:50, every word sung in both
Duet
Takes of 1:03 and 1:00, every word sung in both
Rap
Takes of 0:40 and 0:51, every word sung in both

The rap voice moved fastest and finished first; the duet took the most room. Length also changed between two takes with the same voice, by 19 seconds for the female one, because each take is a performance of its own. Eight lines make a short song whichever voice sings them: our 159-word lyric, two verses, a bridge and a chorus sung twice, ran about two and a half minutes.

Basis: songs we made on Lyricsung on 2 November 2026 from 22 lyrics we wrote for testing, two takes each, with the settings shown. “Every word” and the scores come from the engine’s own word check of each take, not from listening.

Will it say names and numbers right?

Not always. A singer reading letters can’t know that Aoife is said “Eefa” or that 6:45 is a time, so these are the words to check first. We wrote one song with them twice: the first verse the way you would type it (Aoife and Tadhg, Dr. Wojcik, the 6:45 out of NYC, ’09, 24/7), the second spelled the way it is said (Eefa and Tige, Doctor Voychick, the six forty-five out of N Y C, oh-nine, twenty-four seven).

Both takes sang every line. But the engine’s own word check, which scores out of 100 how closely what was sung matches what was written, put this song last of our 22: 72 and 78, where every other song scored between 87 and 97. The score can’t say which verse went wrong, since odd spellings confuse the check as much as the singer, so treat it as a warning and listen.

  • Write a name the way it is said when that differs from the spelling, and listen to the chorus first, where it comes back.
  • Write numbers as words: “six forty-five” rather than “6:45”, and a year the way you say it.
  • Spell short forms out (Doctor, Saint, Avenue), or space the letters (N Y C) when you want the letters sung.

Lyrics in Spanish, French and German

We wrote a short song in each, “Quédate”, “Reste encore un peu” and “Bleib noch ein bisschen hier”, and sang them in the pop style without choosing a language: the words decide it. All six takes sang every word, and the word check scored them 88 to 92, in line with our English songs at 87 to 97.

That shows the words came through clearly enough to be recognised. It doesn’t show how natural the accent sounds to someone who grew up with the language; if a take sounds off to you, run it again, because the next take is a new performance.

Basis: songs we made on Lyricsung on 2 November 2026 from 22 lyrics we wrote for testing, two takes each, with the settings shown. “Every word” and the scores come from the engine’s own word check of each take, not from listening.

Questions about the singer

Which AI can sing my lyrics?

Any lyrics-to-song tool built on an AI music model will sing them; here you choose a voice above and paste them in. Before you pick a tool, check three things: whether it keeps your words as written, how long a lyric it takes, and what you may do with the song. On Lyricsung the box takes 5,000 characters, and songs are for personal use for now.

Can I hear just the voice?

Yes, on Pro or Ultra. Stems, in a song’s More menu, splits it into two files, the vocals and the backing, for one credit; a split you have already made comes back free.

Can I use my own voice?

No. Lyricsung sings with AI voices only: it can’t learn your voice from a recording, and it won’t copy anyone else’s.

The voices and the lyrics box are at the top of this page. For a lyric that isn’t ready yet, our guide to preparing lyrics has a checklist.

Choose a voice