[GH-ISSUE #2872] RVC training fails because mute40k.wav is missing, but the WebUI reports success #2073

New issue

Open

opened 2026-10-23 21:11:03 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-23 21:11:03 +00:00

Owner

Originally created by @pilarblancom-max on GitHub (Oct 21, 2026).
Original GitHub issue: https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/issues/2872

Hello,

I am trying to train a voice model using RVC WebUI on Windows 11, but I cannot complete the training successfully.

My setup

Windows 11
Python 3.12.2
PyTorch 2.4.1 (CPU) and torch-directml
AMD Ryzen 9 4900H with integrated AMD Radeon graphics
RVC WebUI, v2 model, 40 kHz, with pitch guidance enabled
Experiment name: MiVoz

I have three WAV recordings of my own voice, approximately nine minutes in total.

What works

Dataset preprocessing completes successfully.
F0 extraction and HuBERT feature extraction complete successfully.
The feature extraction log reports 45 successful F0 extractions and no errors.
The index is created successfully as added_IVF202_Flat_nprobe_1_MiVoz_v2.index.

However, no model checkpoint (.pth) is created in either logs/MiVoz or assets/weights. Consequently, the model does not appear in the voice selection dropdown.

The error in the console

The training process terminates with this traceback:

FileNotFoundError: [WinError 3] The system cannot find the path specified:
'C:\Users\pilar\Downloads\Retrieval-based-Voice-Conversion-WebUI-main\Retrieval-based-Voice-Conversion-WebUI-main/logs/mute/0_gt_wavs/mute40k.wav'

The WebUI nevertheless reports all four stages as completed, including model training.

There is no logs/mute directory in my installation. I also checked the subdirectories manually and confirmed that no model .pth files were generated.

In webui.py, I found code that adds paths to logs/mute/0_gt_wavs/mute40k.wav and related files to the training file list. However, I have not found where these files are supposed to be created.

During troubleshooting, I also encountered an import error involving train/train.py. I modified the subprocess commands in webui.py to run the script with Python's -P option, and restored the import in train/train.py to:

from train import utils

Dataset preprocessing and feature extraction now work, but the training process still fails with the missing mute40k.wav error.

My questions

Should RVC automatically generate the files under logs/mute, or must they be downloaded or created separately?
If they need to be generated, what is the correct procedure for this version of RVC?
Could the changes to the subprocess commands or my Python 3.12 environment be contributing to the problem?
How can I fix this without repeating the preprocessing and feature extraction stages unnecessarily?

I would appreciate any advice. Thank you for your time and help.

Originally created by @pilarblancom-max on GitHub (Oct 21, 2026). Original GitHub issue: https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/issues/2872 Hello, I am trying to train a voice model using RVC WebUI on Windows 11, but I cannot complete the training successfully. **My setup** - Windows 11 - Python 3.12.2 - PyTorch 2.4.1 (CPU) and torch-directml - AMD Ryzen 9 4900H with integrated AMD Radeon graphics - RVC WebUI, v2 model, 40 kHz, with pitch guidance enabled - Experiment name: `MiVoz` I have three WAV recordings of my own voice, approximately nine minutes in total. **What works** - Dataset preprocessing completes successfully. - F0 extraction and HuBERT feature extraction complete successfully. - The feature extraction log reports 45 successful F0 extractions and no errors. - The index is created successfully as `added_IVF202_Flat_nprobe_1_MiVoz_v2.index`. However, no model checkpoint (`.pth`) is created in either `logs/MiVoz` or `assets/weights`. Consequently, the model does not appear in the voice selection dropdown. **The error in the console** The training process terminates with this traceback: ```text FileNotFoundError: [WinError 3] The system cannot find the path specified: 'C:\Users\pilar\Downloads\Retrieval-based-Voice-Conversion-WebUI-main\Retrieval-based-Voice-Conversion-WebUI-main/logs/mute/0_gt_wavs/mute40k.wav' ``` The WebUI nevertheless reports all four stages as completed, including model training. There is no `logs/mute` directory in my installation. I also checked the subdirectories manually and confirmed that no model `.pth` files were generated. In `webui.py`, I found code that adds paths to `logs/mute/0_gt_wavs/mute40k.wav` and related files to the training file list. However, I have not found where these files are supposed to be created. During troubleshooting, I also encountered an import error involving `train/train.py`. I modified the subprocess commands in `webui.py` to run the script with Python's `-P` option, and restored the import in `train/train.py` to: ```python from train import utils ``` Dataset preprocessing and feature extraction now work, but the training process still fails with the missing `mute40k.wav` error. **My questions** 1. Should RVC automatically generate the files under `logs/mute`, or must they be downloaded or created separately? 2. If they need to be generated, what is the correct procedure for this version of RVC? 3. Could the changes to the subprocess commands or my Python 3.12 environment be contributing to the problem? 4. How can I fix this without repeating the preprocessing and feature extraction stages unnecessarily? I would appreciate any advice. Thank you for your time and help.