### Start Training

Source: https://github.com/aslp-lab/meanvc/blob/main/README.md

Initiate the model training process. Ensure configurations in `script/train.sh`, such as data paths and model directory, are correctly set.

```bash
bash script/train.sh
```

--------------------------------

### Clone Repository and Set Up Environment

Source: https://github.com/aslp-lab/meanvc/blob/main/README.md

Clone the MeanVC repository, create a Conda environment, and install necessary dependencies using the provided script.

```bash
git clone https://github.com/ASLP-lab/MeanVC.git
cd MeanVC

conda create -n meanvc python=3.11 -y
conda activate meanvc

pip install -r requirements.txt
```

--------------------------------

### Download Pre-trained Models

Source: https://github.com/aslp-lab/meanvc/blob/main/README.md

Execute this script to automatically download the main VC model, vocoder, and ASR model. The speaker verification model requires manual download and placement.

```bash
python download_ckpt.py
```

--------------------------------

### Real-Time Voice Conversion

Source: https://github.com/aslp-lab/meanvc/blob/main/README.md

Run this script to perform real-time voice conversion using your microphone as input and a specified target voice. Ensure the target voice path is correctly provided.

```bash
python src/runtime/run_rt.py --target-path "path/to/target_voice.wav"
```

--------------------------------

### Prepare Data List for Training

Source: https://github.com/aslp-lab/meanvc/blob/main/README.md

Format for the training data list file. Each line specifies utterance ID, paths to content features, Mel spectrograms, speaker embeddings, and optional prompt Mel spectrograms.

```text
# Format: utt|bn_path|mel_path|xvector_path|prompt_mel_path1|prompt_mel_path2|...
utterance_id_001|/path/to/bns/utt001.npy|/path/to/mels/utt001.npy|/path/to/xvectors/utt001.npy|/path/to/mels/prompt01.npy
```

--------------------------------

### Data Preprocessing: Extract Content Features

Source: https://github.com/aslp-lab/meanvc/blob/main/README.md

Extract content features (BN) using a 160ms window. Specify input and output directories.

```python
python src/preprocess/extract_bn_160ms.py --input_dir path/to/wavs --output_dir path/to/bns
```

--------------------------------

### Data Preprocessing: Extract Mel Spectrograms

Source: https://github.com/aslp-lab/meanvc/blob/main/README.md

Extract Mel spectrograms from audio files with a 10ms frame shift. Specify input and output directories.

```python
python src/preprocess/extrace_mel_10ms.py --input_dir path/to/wavs --output_dir path/to/mels
```

--------------------------------

### Data Preprocessing: Extract Speaker Embeddings

Source: https://github.com/aslp-lab/meanvc/blob/main/README.md

Extract speaker embeddings using WavLM. Specify input and output directories.

```python
python src/preprocess/extract_spk_emb_wavlm.py --input_dir path/to/wavs --output_dir path/to/xvectors
```

--------------------------------

### Offline Voice Conversion

Source: https://github.com/aslp-lab/meanvc/blob/main/README.md

Use this script for batch processing or converting pre-recorded audio files. Configure source, reference, and output paths within the script before execution.

```bash
bash scripts/infer_ref.sh
```

--------------------------------

### MeanVC Project Citation (BibTeX)

Source: https://github.com/aslp-lab/meanvc/blob/main/README.md

Use this BibTeX entry to cite the MeanVC paper in academic publications. Ensure proper formatting when including in your bibliography.

```bibtex
@article{ma2025meanvc,
  title={MeanVC: Lightweight and Streaming Zero-Shot Voice Conversion via Mean Flows},
  author={Ma, Guobin and Yao, Jixun and Ning, Ziqian and Jiang, Yuepeng and Xiong, Lingxin and Xie, Lei and Zhu, Pengcheng},
  journal={arXiv preprint arXiv:2510.08392},
  year={2025}
}
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.