README.md

---
license: cc-by-nc-sa-4.0
language:
- en
- zh
task_categories:
- audio-text-to-text
- question-answering
pretty_name: AudioSpan
size_categories:
- 1K<n<10K
tags:
- audio
- long-form-audio
- audio-comprehension
- benchmark
- audio-question-answering
configs:
- config_name: accuracy
  default: true
  data_files:
  - split: S
    path: metadata/accuracy/S.jsonl
  - split: M
    path: metadata/accuracy/M.jsonl
  - split: L
    path: metadata/accuracy/L.jsonl
- config_name: rubric
  data_files:
  - split: S
    path: metadata/rubric/S.jsonl
  - split: M
    path: metadata/rubric/M.jsonl
  - split: L
    path: metadata/rubric/L.jsonl
- config_name: chain
  data_files:
  - split: S
    path: metadata/chain/S.jsonl
  - split: M
    path: metadata/chain/M.jsonl
  - split: L
    path: metadata/chain/L.jsonl

---

# AudioSpan: Spanning the Duration and Depth of Audio Comprehension

<p align="center">
  <a href="https://arxiv.org/abs/2608.26431"><img src="https://img.shields.io/badge/Paper-arXiv-b31b1b?style=flat-square&logo=arxiv&logoColor=white" alt="arXiv"></a>
  <a href="https://huggingface.co/datasets/holvan/AudioSpan"><img src="https://img.shields.io/badge/Dataset-AudioSpan-ffd21e?style=flat-square&logo=huggingface&logoColor=000" alt="Hugging Face Dataset"></a>
  <a href="https://creativecommons.org/licenses/by-nc-sa/4.0/"><img src="https://img.shields.io/badge/License-CC%20BY--NC--SA%204.0-3da639?style=flat-square" alt="License: CC BY-NC-SA 4.0"></a>
</p>


## Introduction

**AudioSpan** is a benchmark for **long-form audio comprehension**, spanning
diverse durations and cognitive depths.

<p align="center">
  <img src="figs/audio_span.png" alt="AudioSpan overview" width="100%">
</p>


Questions come from two complementary paths:

- **Native QA**: questions drawn from the audio's natural content.
- **Anchor QA**: questions built around acoustic anchors planted into the
  audio.

Each path is scored in its own mode:

- **Accuracy**: multiple choice questions on native audio, scored by exact
  match.
- **Rubric**: open-ended questions on native audio, graded by rubric-based
  LLM judges against criteria.
- **Chain**: multiple-choice question chains on anchor audio; an answer is
  credited only up to the first error in the chain.

## Quick Start

Download the release (~18 GB) and work from its root:

```bash
pip install -U huggingface_hub
hf download holvan/AudioSpan --repo-type dataset --local-dir AudioSpan
cd AudioSpan
```

### 1. Prepare Audio Data

Native recordings and sound events ship as tar archives
(`audio/audio_native.part*.tar`, `audio/audio_events.tar`; the native parts
are concatenated automatically). From the release root (the directory with
`prepare/` and `audio/`), verify them against `audio/CHECKSUMS.sha256` and
unpack into `audio/`:

```bash
# accuracy / rubric: native recordings only
python prepare/unpack_audio.py --buckets native
# -> audio/native/*.flac

# chain: also rebuild the anchor audio from the native recordings and the
# exact edit recipes in metadata/media/anchor/anchor_manifest.jsonl
# (needs only ffmpeg)
python prepare/unpack_audio.py --buckets native,events
python prepare/prepare_anchor.py
# -> audio/anchor/*.flac
```

(`unpack_audio.py` resolves the release root from its own location; pass
`--root` if you run it from elsewhere.)

### 2. Prepare Inference Results

The questions live in `metadata/{accuracy,rubric,chain}/{S,M,L}.jsonl`; each
record carries the question, its `audio_path`, and (for multiple choice) the
four options. Run your model over each record and write one answer per
question — only `qa_id` and the model's output; ground truth stays in the
metadata and is joined in by the scorer:

**accuracy / chain** (multiple choice):

```json
{"qa_id": "S_EN_001_P", "answer": "D"}
```

**rubric** (open-ended):

```json
{"qa_id": "S_EN_001_P", "answer": "The narrator first says HDR at about 06:04 ..."}
```

- `answer` — the model's raw output; for multiple choice the scorer
  extracts the option letter (A/B/C/D) from it.
- Questions missing from your file count as wrong (or score zero).

### 3. Run Evaluation

```bash
# accuracy + chain: stdlib-only, no dependencies
python evaluate/score.py --mode accuracy --input your_accuracy.jsonl
python evaluate/score.py --mode chain    --input your_chain.jsonl

# rubric: needs an LLM judge (default: gpt-5.4-2026-03-05; API key from the
# matching provider env var, e.g. OPENAI_API_KEY; or pass --api-base/--api-key)
pip install openai tenacity
export OPENAI_API_KEY="your-api-key-here"
python evaluate/score_rubric.py --input your_rubric.jsonl
```

Both scorers join your answers against the ground truth in `metadata/` by
`qa_id` — no reference answers needed in your submission.

## License and Data Use

AudioSpan is released for **non-commercial research and evaluation**. Our
artifacts (QA items, rubric criteria, anchor manifests, sound events, and
evaluation scripts) are under CC BY-NC-SA 4.0; copyright of the source
recordings remains with their original creators. The recordings are
included only for evaluation, and downloading the dataset constitutes
agreement not to redistribute the audio or use it in commercial products.

## Citation

If you find AudioSpan useful for your research, please consider citing:

```bibtex
@article{huang2026audiospan,
  title   = {AudioSpan: Spanning the Duration and Depth of Audio Comprehension},
  author  = {Huang, Wen and Chu, Yunfei and Gao, Meng and He, Haolin and Xu, Jin},
  journal = {arXiv preprint arXiv:2608.26431},
  year    = {2026}
}
```