RVC
RVC sings a vocal stem again in one of four packaged voices. It is the higher quality of the two voice conversion options in Miso, and the less flexible one.
What it does in Miso
Section titled “What it does in Miso”One task, Convert the voice, on the remix page.
The four voices
Section titled “The four voices”default, manthos, chocola and fraise.
Miso checks your choice against that list before the job queues. Sent to the backend, an unknown voice is refused only after the job has queued and the weights have loaded, and the list is four entries long, so checking it up front costs nothing.
A user .pth model file is supported by the runtime. Miso does not expose that.
Options
Section titled “Options”| Option | Does |
|---|---|
pitch_extractor |
Standard is rmvpe |
semitone_shift |
Pitch shift in semitones |
retrieval_blend |
Blend rate with the FAISS index |
RVC is the one task in Miso whose options travel nested, inside an options object rather than beside the audio. This is worth knowing if you are working on Miso itself: sent flat they are accepted and ignored, and the run comes back with a perfectly good track that heard none of them.
The rule mirrors the CLI, where a flag is a top level field and a request option is not.
Sample rates
Section titled “Sample rates”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.
Where the result lands
Section titled “Where the result lands”Beside the stems it was converted from, not in the song list, and with the chosen voice added to its name. Two conversions of the same stem therefore do not end up with identical names.
RVC or Vevo2
Section titled “RVC or Vevo2”| RVC | Vevo2 | |
|---|---|---|
| Voices | Four packaged | Any track in the project |
| Output | 40 kHz | 24 kHz mono |
| Quality | Higher | Lower bandwidth |
The two sit beside each other rather than one replacing the other. RVC sounds better. Vevo2 can use any voice.