Changing a voice

Two tasks change who is singing. They sit beside each other rather than one replacing the other, because they trade different things away.

	RVC	Vevo2
Voices	Four packaged	Any track in the project
Output	40 kHz	24 kHz mono
Needs	A vocal stem	A vocal stem and a reference track

RVC sounds better. Vevo2 can use any voice. Pick accordingly.

RVC: four packaged voices

Convert the voice sings a vocal stem again in one of four voices: default, manthos, chocola and fraise.

Run it from the remix page on a vocal stem. The result lands beside the stems it was converted from rather than in the song list, and the chosen voice is added to its name so two conversions of the same stem do not end up with identical names.

RVC answers at 40 kHz whatever it is given. Miso converts the result back to the rate its source arrived at, because the stems it has to sit beside are 44.1 kHz and the mix route refuses a set whose rates disagree.

It accepts 44.1 kHz and 48 kHz sources alike, so there is no rate you have to fix first.

Vevo2: any voice you have

Sing it in another voice converts a vocal stem to the voice of any other track in the project.

It reads two tracks, so the form has two pickers: the stem being converted, and the track to copy the voice from. A few seconds of clean vocal is enough for the reference.

Everything comes back at 24 kHz mono, on every route. Nothing above 12 kHz survives. That is the trade: bandwidth for being able to use any voice at all.

Under the hood this runs Vevo2’s style_preserved_svc route, which keeps the original performance and changes only who is singing it.

Singing written lyrics

Sing lyrics in a voice is a generator rather than a remix, so it lives on the create page.

Write lyrics, pick a track to borrow the voice from, and it sings them. Add a second track as a melody and it follows that instead of writing its own.

That second track is what decides the route:

• No melody runs text_to_singing. Vevo2 writes its own tune, and the length follows your words.
• With a melody runs humming_to_singing. It follows the tune you gave it, and the length follows the melody.

The length limit on the form is what stops a long lyric short in the first case. The upstream default of 500 tokens is about seven seconds, which is why Miso sends 1500.

Output is 24 kHz, the same as every other Vevo2 route.

Getting a vocal stem

All three of these want a vocal stem, not a whole song. Split the track first.