XiaomiMiMo/MiMo-Audio-Evalset

---
license: mit
---

# Dataset Card for MiMo-Audio-Evalset

## Dataset Description

This repository is a collection of multiple audio datasets used for evaluation in the [MiMo-Audio-Eval](https://github.com/XiaomiMiMo/MiMo-Audio-Eval) toolkit. It includes a variety of datasets for tasks such as automatic speech recognition (AISHELL1, LibriSpeech), text-to-speech (SeedTTS), audio understanding and reasoning (MMAU and MMSU), and more.

## Included Datasets

The following datasets are included in the MiMo-Audio-Evalset:

* [AISHELL1](https://www.aishelltech.com/kysjcp)
* [LibriSpeech](https://www.openslr.org/12)
* [SeedTTS](https://github.com/BytedanceSpeech/seed-tts-eval)
* [MMAU](https://sakshi113.github.io/mmau_homepage/)
* [MMSU](https://huggingface.co/datasets/ddwang2000/MMSU)
* [ESD](https://github.com/HLTSingapore/Emotional-Speech-Data)
* [Expresso](https://speechbot.github.io/expresso/)
* [MultiChallenge](https://scale.com/research/multichallenge): Speech version of the `MultiChallenge` dataset.
* Speed Conversion: A self-constructed dataset for speech speed conversion.

## Dataset Structure

For each dataset mentioned above, there is a corresponding `.tar.gz` compressed archive. You can download these archives, then extract them to the `data` directory for use with the evaluation toolkit. The expected local directory structure after extraction should be as follows:

```
├── ESD
├── MMAU
├── MMSU_Bench
├── aishell1
├── expresso
├── librispeech
├── multi_challenge
├── seedtts
└── speed_conversion
```

## Download and Extract Script

Here’s an example script to download and extract the necessary `.tar.gz` files:

```python
import os
import tarfile
from huggingface_hub import snapshot_download
from pathlib import Path

data_dir = Path("data")
data_dir.mkdir(parents=True, exist_ok=True)

snapshot_download(
repo_id="XiaomiMiMo/MiMo-Audio-Evalset",
repo_type="dataset",
local_dir=data_dir,
local_dir_use_symlinks=False
)

for tar_file in data_dir.glob("*.tar.gz"):
with tarfile.open(tar_file, "r:gz") as tar:
tar.extractall(path=data_dir)
print(f"Extracted {tar_file.name}")
```

## Disclaimer

We do not own the copyright for these datasets and models. Please adhere to the usage terms and licenses provided by the original dataset and model creators. Ensure compliance with their terms of use when utilizing this data.

## Contact

For any questions or further details about the dataset, please contact us at [[e-mail]]([e-mail]).