Make an AI voice model file
PTH file converter online: turn audio into an RVC voice model
“PTH file converter” means two different jobs, so start with the one you need. Making a .pth from a voice recording is training, and that is this page. Using a .pth you already have to convert a vocal is inference - that happens on the RVC voice converter.
A PTH file converter for RVC actually trains a voice model rather than re-encoding a file. Upload a voice recording and NiceVois runs the complete RVC v2 training in the cloud, then hands back a .pth, its matching .index, and a ZIP of both.
People searching for a “PTH file converter” usually want one thing: a voice model file of their own, without installing a training stack. That is exactly what this does. The difference is that real training happens in between, which is why it takes minutes rather than seconds.
What a .pth file actually is
A .pth is a PyTorch checkpoint holding the trained weights of an RVC voice model. It is the voice. You cannot play it like audio; you load it into an RVC interface, which then converts singing or speech into that voice.
That is also why no tool can convert an MP3 into a .pth the way it converts MP3 to WAV. One is a recording, the other is a trained model. The audio is the teaching material, not the source format.
Your models are files you own
NiceVois gives you the actual files: the .pth, the matching .index, and a ZIP of both. Download them once and no platform can take them away. They keep working in any compatible RVC interface, whether or not NiceVois still exists.
NiceVois keeps a copy available for seven days so you can download it again in that window. After that, the copy you saved is the one that matters, so archive the ZIP somewhere safe.
How the training works, and what makes a model good
Training plays your recording to the model again and again. Each complete pass is one epoch, and with every pass the model moves closer to the timbre and delivery of that voice. That is why it takes minutes rather than seconds, and why the recording you pick matters more than any setting.
Clean beats long. A few clean minutes of one voice, with no music, echo, or second speaker, will beat an hour of messy audio. Variety helps as well, because a recording that moves through different pitches and volumes gives the model more of the voice to learn.
More epochs is not automatically better. Past a point the model starts copying the quirks of your particular recording instead of learning the voice, which comes out harsh or brittle. Around 100 to 200 epochs suits most recordings.
What you need to make one
- A recording of one voice. Two to fifteen minutes is the recommended range, not a rule.
- Little or no background music, echo, or overlapping speakers.
- WAV, MP3, FLAC, M4A, or OGG. A clean recording matters far more than the extension.
- Permission to use the voice. NiceVois trains voices you own or have clear permission to use.
What you get back
| File | What it holds | Why you keep it |
|---|---|---|
| .pth | The trained voice model weights | The voice itself. Load this into any compatible RVC interface. |
| .index | Retrieval index from the training features | Improves similarity to the source voice where the interface supports it. |
| Complete ZIP | Both files together | The safest single thing to archive or move between machines. |
What it costs
Your first training is free. Every job receives a fixed quote from audio length and epochs before it starts, and a subscription covers training, conversion, separation and cleanup with no limits.
Upload a supported recording, name the model, and download the finished files.
Open the converterPTH file questions
What is a .pth file in AI voice?
A .pth file holds the trained weights of an RVC voice model. It is the voice itself, saved as a PyTorch checkpoint, and it is what you load into an RVC interface to convert singing or speech into that voice. It is not audio and cannot be played.
Can you convert an audio file to a .pth file?
Not by file conversion. A .pth is produced by training, not by re-encoding. The audio teaches a model what the voice sounds like, and the finished weights are saved as .pth. NiceVois runs that training in the cloud, so uploading audio and receiving a .pth works like a converter even though real training happens in between.
How do I make my own .pth voice model file?
Upload your audio, name the model, choose the epochs, and press Start Training. NiceVois runs the complete RVC v2 workflow and returns a .pth, its matching .index, and a ZIP of both. Any length of audio will train. Two to fifteen minutes of clean, single-speaker audio is what we recommend for a dependable voice. Your first run is free with a free account.
Do I keep the .pth file?
Yes. The .pth, .index, and ZIP are real files you download to your own device, and they keep working in any compatible RVC interface. NiceVois keeps a copy available for seven days so you can download it again during that window, so save the ZIP somewhere safe once it is ready.
What is the difference between .pth and .index?
The .pth holds the trained voice. The .index is a retrieval index built from the same training features, used by RVC interfaces that support feature retrieval to make output closer to the source voice. Keep both together.