Which AI singer can sing your own lyrics, and how do you pick the voice?

You have a chorus saved on your phone. Eight lines, a working title, maybe a rough sense of how fast it should move. You search "AI singer", open the first three results, and each one asks you for something different: one wants a short description of a style, one opens a piano roll and waits for you to draw notes, and one asks you to upload a recording of somebody singing. All three call themselves AI singing tools. Only one of them can do anything with the eight lines you actually have.

That mismatch is where people lose a weekend, and sometimes a hundred dollars on a licence they cannot use yet. The tools are not bad. They are three different machines sold under one name, and which one you need depends on a single question you can answer before you open any of them.

Three machines sold under one name

The third column is the whole decision. A full-song generator invents everything you did not specify. A vocal synthesizer invents nothing about the melody and sings precisely what you wrote. Voice conversion invents nothing at all: it repaints a performance that already exists.

If you already know your answer is Suno and want the field-by-field handoff, turning a finished lyric into a song walks through what goes in Style, what goes in Lyrics and why stray words end up sung. That page is for the reader who has picked the tool. This one is for the step before: working out which kind of singer your lyric needs, and then how to choose its voice.

The question that sorts you: is there a melody?

Not "can you sing". Not "do you read music". Only this: is there a tune, anywhere, that belongs to these words?

• No tune, just words. You need something that composes and sings in one go. That is a full-song generator. Every other route stalls at the moment it asks you for notes.
• A tune in your head, on a MIDI keyboard or in sheet music. You want that tune sung exactly as written. That is a vocal synthesizer. A full-song generator will not follow your melody, because its form has no field where a melody goes.
• A recording of the song, sung by someone. You, a friend, a guide vocal from a demo. You want it in a different voice. That is voice conversion.

Two mixed cases come up often. A songwriter who has a melody but no patience for editing syllables note by note can still use a full-song generator and accept that the model's tune will replace theirs; sometimes it is better, more often it is simply different. A producer who wants a particular vocal line over a finished beat usually ends up combining routes, and the trade-offs of each for that job are laid out in the piece on AI vocalists for producers.

Full-song generators: words in, finished track out

Suno is the tool this site's generators are built around, so it is the one described here in detail. In Custom mode the create form has a space for your own lyrics, separate from the style description; Suno's help centre documents this. You paste the words, write a short line that describes the genre and the voice, and the model returns a complete track with a melody it composed for those words.

What you control: the words, the structure (through section tags such as [Verse] and [Chorus]), and the character of the voice, through descriptors in the style line. What you do not control: the exact melody, the exact phrasing, and whether the second take sounds anything like the first. Each generation is a new performance. That is useful when you want options, and a nuisance when you finally hear the right one and need a matching second verse.

Two things to check before you plan around Suno.

Rights. Suno's help centre says you keep ownership of original lyrics you type in. At the time of writing its pricing page lists commercial use for the paid plans and not for the free one. Plans change, so read the current page before you release anything.

Voices. Suno now has a feature called Voices that lets you record or upload your own singing (Suno asks for a clip between 15 seconds and 4 minutes, and works best with an a cappella take), checks it against a short phrase you read aloud, and then sings new songs in that voice. Suno's help page says the verification exists to stop people using someone else's voice, that the feature is limited to users aged 18 and over, and that it is not available in some locations.

That tells you something about the "pick the voice" half of the question. In a full-song generator you either describe a voice in words or, with Voices, use your own. You do not get to pick a named human singer, and the platform is built to stop you trying.

Vocal synthesizers: you write the tune, it sings it exactly

This category is older than the full-song generators and works the other way round. You open a piano roll, draw or import notes, type a syllable onto each note, choose a voice and press play. The software sings your melody on your timing. Nothing is invented. When a line sounds wrong, it is because a note or a syllable is wrong, and you fix it by hand: pitch curves, vibrato, the length of a consonant.

Three names come up again and again.

Synthesizer V Studio 2 Pro, from Dreamtonics. At the time of writing it is sold as a one-time purchase, $99 with one voice of your choice included, further official voices at $79 each, and a free trial that falls back to limited functionality when it ends. Its voices sing natively in English, Japanese, Mandarin, Cantonese and Korean, and a cross-lingual feature lets any voice sing in all six of its languages, the sixth being Spanish. Dreamtonics says its official voices come with commercial use licences and are built from recordings of singers who consented.

ACE Studio. It takes MIDI and lyrics and returns a sung vocal, from a library its documentation puts at 140+ voices across eight languages: English, Chinese, Japanese, Korean, Spanish, Italian, French and Portuguese. Beyond the singer it bundles a voice changer and a stem splitter. You can start for free; the current tiers are on the vendor's pricing page.

VOCALOID, from Yamaha. Its site lists voice banks such as Hatsune Miku, and the current editor at the time of writing is VOCALOID6, with VOCALOID7 announced for November 2026. The contract is the same as the others: give it a melody and lyrics, and a voice bank sings them.

Look at what is missing from those language lists. None of the three vendors names Russian among its singing languages at the time of writing, and languages outside each list are a gamble. If your lyric is in a language the vendor does not name, the free trial will answer the question faster than any review.

The detailed side-by-side of the three, editions and languages included, lives in Synthesizer V, ACE Studio or Vocaloid. Read that one if you already know you want a synthesizer and are choosing which.

Voice conversion: same performance, different throat

Voice conversion does not read lyrics at all. It takes audio of someone singing and re-renders it in another voice, keeping the timing, the melody and the phrasing of the original take.

RVC (Retrieval-based Voice Conversion) is an open-source project under the MIT licence. Its repository says a usable model can be trained on ten minutes or less of clean voice data, and everything it does starts from existing audio. It has no mode that turns typed text into singing.

Kits AI packages the same idea as a service: a library of voices it says are licensed from the artists themselves and royalty-free, a tool to build a model of your own voice, a free plan without a card, and paid plans that started at $10 a month at the time of writing.

The catch is the input. Conversion keeps everything about the performance except the timbre. A rushed syllable stays rushed. A flat note stays flat in the new voice. So conversion suits people who already have a good sung take, or who sing in tune and simply dislike their own tone. For someone whose only asset is a page of words, it is the wrong machine, because there is nothing yet to convert.

Conversion is also where rights questions get sharpest, since the obvious temptation is a famous voice. Your own voice, a consent-based library, or a voice you hold written permission for are the routes that keep you out of that argument. This is a description, not legal advice, and the position differs between countries; who owns AI-generated music and our disclaimer set out what this site does and does not claim.

How to choose the voice, route by route

In a full-song generator, the voice is a line of words

Suno has no menu of named singers. The voice is whatever the style line describes, which is why a vague style line is the most common reason a track "sounds wrong". Name the gender, the texture and the delivery, and add the register when it matters to the song. Leave something out and the model fills the gap with its own average.

Two habits make the voice stick. Put the vocal descriptors near the front of the style line, and put the opposite gender in the Exclude field instead of writing "no male vocal" in Style, where a negative tends to be read as a topic rather than a ban. The free guides cover holding one voice across a whole song and stopping a vocal from screaming when the chorus gets loud.

In a vocal synthesizer, match the range before the tone

Here you do pick a named voice bank, and vendors tag them by genre and character. Dreamtonics, for example, lists Felicia under soul, pop, rock and opera, and Hayden under pop and soul. Tone is the easy part. The usual mistake is falling for a voice whose comfortable range sits well away from your melody, then spending an evening fighting thin top notes or mumbled low ones. Find the highest and lowest note of your chorus first. Then audition voices on that chorus during the trial (Dreamtonics gives each voice a 7-day free trial at the time of writing), not on the vendor's demo song, which was written to flatter the voice.

In voice conversion, match the source to the target

The converted voice inherits the range of whoever sang the take. If the guide singer and the target voice sit far apart, expect to transpose and to listen hard for artifacts. Choose a target voice that lives near the range of the take you have, or record the take again in the target's range.

Every route starts with the same thing: the lyric

The three machines share one input. The full-song generator sings your words, the synthesizer places your words on notes, and the conversion tool repaints someone singing your words. If the words do not scan, no singer on this page fixes that.

That part is what this site does. sunomarket.com is an independent site: it is not Suno and is not affiliated with Suno or with any other company named here. The song generator takes a short brief through a form and returns a title and a full lyric with section tags, ready to paste into Suno's Lyrics field. It does not sing. There is no audio at any point. The form has a vocal field with three settings (unspecified, female or male), so the words can be shaped for the voice you are about to choose, and you can pick a structure, a hook type and the language of the lyric.

Here is the shape a finished lyric takes, with delivery notes sitting inside the brackets:

[Verse | soft female vocal, close]
Your coat still hangs behind the door
I move it when I sweep the floor
The pocket holds a ticket stub
For a train we never took

[Pre-Chorus]
And when the kettle starts to sing
I set out two of everything

[Chorus | full voice]
Leave the light on in the hall
Leave it burning, that is all
If you ever find the way back home
You won't walk in from the dark alone

[Bridge | whispered]
I know, I know
The door's been shut a year or so

[Outro | fade out]

Everything inside square brackets is an instruction and everything outside them gets sung; that is Suno's convention. The block above carries the same lyric to all three machines, with one adjustment each. In Suno it goes into the Lyrics field as it stands. In a vocal synthesizer the brackets mean nothing, so you delete them and put each syllable on a note. In voice conversion the lyric is the sheet your guide singer reads from, and the bracket notes become directions for that singer.

On cost: one lyric on the SM1 model is 20 credits, which is $0.20. SM2 and SM3 add an editing pass in which a detector flags weak lines and a repair step rewrites them; they cost 80 credits, $0.80 a lyric. New accounts start at zero credits and there is no free trial of the lyric generator. The plans are $5 a month for 500 credits, or $19 once for 1,500 credits that never expire. What is free without a card: the full recipes for the ten most popular styles in the catalogue and every prompt guide.

Choose a route in five minutes

Before paying for anything, answer the melody question out loud. Then run the one free test that matches your answer: a Suno generation with your real lyric, a Synthesizer V or ACE Studio trial on your actual chorus rather than the demo, or a short conversion of a take you already have. And if the lyric itself is the thing you do not have yet, start there, because every singer above is waiting on it.