NNiceVois

Lossless audio input

FLAC to PTH: train an RVC model from lossless audio

Upload a FLAC and start trainingLossless in, portable .pth out.
Summary

Yes, and if you have a choice, FLAC is the one to upload. FLAC is lossless: it compresses the file without discarding any audio, so the model trains on every detail the microphone captured. An MP3 or M4A of the same recording has already thrown some of that away permanently.

That said, format is the smallest of the three things that decide model quality. A noisy FLAC will lose to a clean MP3 every time. Lossless only helps when the recording underneath it is already good.

Voice training

Train an RVC voice from an audio file

Add up to 50 WAV, MP3, FLAC, M4A, or OGG files. NiceVois keeps them separate and prepares them together as one private training dataset.

.pth

This name is used for the .pth, .index, and complete ZIP package.

Standard is selected automatically.
200 epochs

Recommended: add your audio and NiceVois will suggest the right number.

Clean audio produces a better model. Use a dry voice recording with minimal music, echo, or background noise. Prepare your audio first if it needs it.

Persistent cloud storage

Your model library

Your voice models are files you own. Download them once and no platform can take them away.

Loading…
Finished model packages stay saved until you delete them.

Your .pth, .index, ZIP, and private listening preview remain available after closing the browser or restarting your computer. Full-resolution source audio and temporary training files are removed after validation.

Loading your models

Checking private cloud storage…

What actually decides quality, in order

Format matters least. Spend your effort in this order:

  1. One speaker, cleanly recorded. Other voices, music, and room noise all become part of what the model learns.
  2. Enough material. Several minutes of varied speech beats a single perfect sentence.
  3. Consistent capture. One microphone, one room, one distance.
  4. Then the format - where FLAC is the best available answer.

When FLAC does not help

Converting a lossy file into FLAC does not restore anything. An MP3 exported to FLAC is still an MP3 underneath - the file grows, the audio does not improve, and the upload simply takes longer. Only use FLAC when it came from a lossless original.

What the service does with the FLAC

The official RVC training workflow reads supported source audio through FFmpeg, decodes it into the representation preprocessing requires, and extracts the features used for training. NiceVois runs that workflow in a pinned cloud environment and validates the exported model before showing the downloads. FLAC needs no preparation on your side.

FLAC versus the alternatives

QuestionFLACMP3 / M4A
Can this tool upload it?YesYes
CompressionLossless - nothing is discardedLossy - detail is permanently removed
Typical sourceStudio capture, archival rip, DAW exportPhone, download, messaging app
File sizeLargeSmall
Best choiceUse whenever you have itUse when it is the cleanest original you own

What you receive

The finished model is available as a named .pth, a matching .index, and a ZIP containing both. The public workspace retains the completed files for the displayed seven-day window, and you can delete an inactive job from your private model library at any time.

Have a lossless recording?

FLAC is the strongest source you can hand a voice model. Upload it directly.

Train a PTH model from FLAC

Technical reference: RVC Project training instructions.