DaFeiYu RVC v2 40k

DaFeiYu RVC v2 40k

Converts a recorded speaking or singing voice into the 大肥鱼 voice, as a 40k RVC v2 model with pitch guidance. It needs an RVC inference tool and is for personal, non-commercial projects.

Made by YukiKaze1337 on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: DaFeiYu RVC v2 (40k) is a voice conversion model for RVC, not a text-to-speech model. It takes a recording of speech or singing and makes it sound like the 大肥鱼 voice. You run it inside an RVC inference tool, with the model file and its index file.

Who it suits: It suits hobby creators who want to convert recorded voices for personal or learning projects. Use only voices you have the right to use.

Industry | Who uses it | How they use it | Example
Music covers | Hobby creators | Convert a recorded vocal line into this voice | A sung line that comes out in the 大肥鱼 voice
Fan and video projects | Non-commercial content makers | Convert a spoken recording for a personal project | A short spoken clip re-voiced for a fan video
Audio learning | People exploring RVC | Study how a small RVC v2 model behaves | Compare a converted file with the original recording

You can skip it if you need a voice that reads typed text aloud, or if your project is commercial.

Where to use it:

• The RVC WebUI, with the two model files copied into its folders
• The command line of the official RVC repository
• Windows, macOS and Linux with Python, with a macOS note under Good to know

Use case | Who | What they get back
Singing voice conversion | Hobby creators | A WAV file of the input vocal in this voice
Speech voice conversion | Content makers | A converted recording of the spoken input
Pitch-matched covers | Singers | An output shifted to suit the source voice

How to set it up: You need these first:

• An RVC inference tool, either the WebUI or the official repository's command line
• Python, to run the tool and the Hugging Face client
• Disk space for two files, a 53 MB weights file and a 154 MB index file

1. Install an RVC inference tool.
2. Download this repository's files with the two commands below.
3. Copy dafeiyu.pth into assets/weights/ and the index file into assets/indices/, or into logs/.
4. Select both in the tool and convert a recording.

pip install -U huggingface_hub
hf download YukiKaze1337/DaFeiYu_rvc_v2_40k

How to use it: In the WebUI, you pick the model and the index at inference time and convert your audio. From the command line of the official repository, you give the model, the input, the output and the index.

python infer/cli.py \
  --model assets/weights/dafeiyu.pth \
  --input input.wav \
  --output output.wav \
  --index assets/indices/added_IVF1314_Flat_nprobe_1_dafeiyu_v2.index \
  --pitch 12 --f0-method rmvpe --format wav

Tool | What it does
infer/cli.py | Runs one conversion from the command line
--model | Points to the dafeiyu.pth weights file
--index | Points to the retrieval index, which helps the result sound closer to the voice
--pitch | Shifts the pitch, with 12 meaning one octave up
--f0-method | Chooses how pitch is detected, here rmvpe

Male source recordings usually work with a pitch of 12, and female ones with 0.

Example: You take a recording of a male voice singing, then run the command with a pitch of 12 and the index file. You get an output.wav of the same line in the 大肥鱼 voice.

For newcomers and experts: RVC changes whose voice a recording sounds like, while keeping the words and the tune. It does not read text. A good first try is a short WAV file run through the WebUI with the model and index selected.

For readers who know RVC, this is a v2 model at 40k with pitch guidance. It reserves 109 speaker slots and uses only speaker 0. The training set was 227 slices of machine-synthesized speech, about 11.1 minutes, taken to roughly 2.33 million steps. The index holds 51,267 features at 768 dimensions. The author says 11 minutes is short, so the tone is steady but high or held notes can drift.

Pick an input recording
Load dafeiyu.pth and the index file
Set the pitch and the f0 method
Run the conversion
Save the output audio

Good to know:

• Training data: machine-synthesized speech, not recordings of a real person, in a separate dataset on Hugging Face.
• macOS: using faiss and torch together with OpenMP can crash, so set OMP_NUM_THREADS=1 before inference.
• Language: Hugging Face lists it for Chinese.

Title: File sizes
Weights file: 53 MB
Index file: 154 MB

• License: MIT.
• Cost to run: none named. It runs on your own computer.
• Data it reaches: the audio files and model files on your computer. The source names no web access.
• Not for: text-to-speech, impersonating anyone, fraud or forgery, or commercial work. The card allows personal, learning and entertainment use.

On Hugging Face, 13 November 2026

• Task: audio to audio
• Library: rvc
• Downloads in the last 30 days: 571
• Likes: 10
• License: mit
• Languages: zh
• Tags: voice-conversion, singing-voice-conversion, audio, audio-to-audio

Links

• Model: https://huggingface.co/YukiKaze1337/DaFeiYu_rvc_v2_40k
• Files: https://huggingface.co/YukiKaze1337/DaFeiYu_rvc_v2_40k/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download YukiKaze1337/DaFeiYu_rvc_v2_40k

Then follow the usage in its model card. Nothing in this kit runs by itself.