Clone a speaking voice to sing — you do not have to perform the song

You can clone a speaking voice and still get a sung cover. That is the default path, not a workaround. The studio asks you to read three short English prompts. It does not ask you to belt the chorus. Do I need to know how to sing? — no.

If you can leave a voicemail, you can start. The singing is the model's job.

What the samples are for

The prompts capture timbre: brightness, weight, accent, how you finish a word. They are not a rehearsal of Anti-Hero. Generation takes the original vocal's melody and timing and rebuilds the sound with that timbre.

So: speaking in, singing out. The longer explanation of that translation — and where it gets thin on held notes and big leaps — is speaking voice vs singing voice conversion.

You still do not perform the full track. One extra slightly musical sample is optional, not a requirement.

Record like a voice memo, not an audition

Open Studio. Quiet room. Same distance from the mic for all three takes. Speak the way you actually talk — not a radio voice, not a whisper.

• Built-in laptop or iPhone mics are enough. Recording on iPhone if that is the device.
• Do not record in Voice Memos and then share a compressed file in. Record in the studio.
• Expression helps more than volume. A flat grocery-list read gives the model less pitch variation to steal from when it has to sing.

Clean samples beat "trying to sound like a singer" for a first train. Free is one lifetime voice training — treat that run as the one you want.

Where speaking-trained clones sound best

Conversational pop and chant hooks hide the speech-to-sing gap. Espresso and APT. are easier first tests than a belted power duet. If the chorus thins out, that is often range the samples never showed — try a mid-range demo with the same model before you assume the clone is broken. When your AI cover sounds robotic.

Another language: you can cover a Korean or Spanish chart track with English speaking samples. You will not become a native singer. Can AI sing in another language?.

What this is not

It is not karaoke scoring. You are not graded on pitch in real time — AI cover vs karaoke and the AI karaoke app guide. It is not text-to-song: the track has to already exist. And it is only your speaking voice. Can I clone someone else's voice? is no.

Credits, then generate

Free: 100 credits a month (about 10 songs), 4 MB uploads, one lifetime train. Train 10, clip 10, cover 15. Pro $9 / 900, Premium $19 / 2,000, Max $39 / 4,200. Pricing.

Then: pick a demo or upload a song you have rights to, generate, open the share page. Friends tap Make your version and read their own prompts. They do not have to sing either.

How to make an AI cover is the end-to-end loop. Can AI clone my voice to sing? is the yes.

Open Studio. Read. Do not audition.