---
configs:
  - config_name: AISHELL1_dev
    data_files:
      - split: test
        path: AISHELL1_dev/metadata.parquet
  - config_name: Background_noise
    data_files:
      - split: test
        path: Background_noise/metadata.parquet
  - config_name: Bilingual_uedin
    data_files:
      - split: test
        path: Bilingual_uedin/metadata.parquet
  - config_name: CommonVoiceFR_dev
    data_files:
      - split: test
        path: CommonVoiceFR_dev/metadata.parquet
  - config_name: Libritts
    data_files:
      - split: test
        path: Libritts/metadata.parquet
  - config_name: Long_context
    data_files:
      - split: test
        path: Long_context/metadata.parquet
  - config_name: Multispeaker_libri
    data_files:
      - split: test
        path: Multispeaker_libri/metadata.parquet
  - config_name: VCTK
    data_files:
      - split: test
        path: VCTK/metadata.parquet
  - config_name: robotcall
    data_files:
      - split: test
        path: robotcall/metadata.parquet
  - config_name: vctk_text_robust
    data_files:
      - split: test
        path: vctk_text_robust/metadata.parquet
---

# RVCBench

RVCBench is a benchmark dataset for studying robustness in voice cloning and related audio generation pipelines.

Each subset is exposed as its own Hugging Face dataset configuration. Most subsets contain:

- `metadata.parquet`
- `audios/`

The canonical metadata stores one row per benchmark pair with columns such as:

- `speaker_id`
- `prompt_file_name`
- `prompt_text`
- `prompt_language`
- `target_file_name`
- `target_text`
- `target_language`
- `pair_id`
- `dataset_name`
- `split`

When available, training-oriented annotations are also preserved:

- `prompt_phonemes`, `prompt_tone`, `prompt_word2ph`
- `target_phonemes`, `target_tone`, `target_word2ph`

Some subsets include additional task-specific metadata, for example `spam_type` in `robotcall`.

## Available Configs

- `AISHELL1_dev`
- `Background_noise`
- `Bilingual_uedin`
- `CommonVoiceFR_dev`
- `Libritts`
- `Long_context`
- `Multispeaker_libri`
- `VCTK`
- `robotcall`
- `vctk_text_robust`

## Notes

- The `compression` directory is intentionally excluded from this dataset release.
- This repository is organized for direct browsing in the Hugging Face dataset viewer.

## Paper

This dataset is introduced in:
[RVCBench: Benchmarking the Robustness of Voice Cloning Across Modern Audio Generation Models](https://arxiv.org/abs/2602.00443)