---
license: cc-by-4.0
task_categories:
  - audio-classification
  - question-answering
language:
  - en
tags:
  - music
  - audio
  - benchmark
configs:
  - config_name: lyrics_transcription
    data_files:
      - split: test
        path: lyrics_transcription/*
  - config_name: artist_collaboration
    data_files:
      - split: test
        path: artist_collaboration/*
  - config_name: structural_segmentation
    data_files:
      - split: test
        path: structural_segmentation/*
  - config_name: musicological_analysis
    data_files:
      - split: test
        path: musicological_analysis/*
---

# BASS: Benchmarking Audio LMs for Musical Structure and Semantic Reasoning

[![Paper](https://img.shields.io/badge/arxiv-PDF-red)](https://arxiv.org/abs/2602.04085)
[![Website](https://img.shields.io/badge/🌐-Website-blue)](https://minjang10.github.io/bass-website)
[![GitHub](https://img.shields.io/badge/GitHub-Code-black)](https://github.com/minjang10/bass_music_benchmark)

[BASS](https://arxiv.org/abs/2602.04085) is a benchmark for evaluating music understanding and reasoning in audio language models. It comprises **2,658 questions** across **12 tasks** and **4 categories**, covering **1,993 unique songs** and over **138 hours of music**.

---

## 🚀 Usage

```python
from datasets import load_dataset

ds = load_dataset("oreva/bass_music_benchmark", "lyrics_transcription")
ds = load_dataset("oreva/bass_music_benchmark", "artist_collaboration")
ds = load_dataset("oreva/bass_music_benchmark", "structural_segmentation")
ds = load_dataset("oreva/bass_music_benchmark", "musicological_analysis")
```

---

## 🗂️ Task Categories

| Category | Description | Tasks |
|---|---|---|
| Structural Lyrics Transcription | Transcribe song lyrics at varying levels of granularity | Full, Section-level |
| Structural Segmentation | Identify and localize structural sections within a recording | Full, Section-level |
| Artist Collaboration | Reason about the presence, duration, and identity of contributing artists | Counting, Duration, Localization, Attribution |
| Musicological Analysis | Assess knowledge of genre attributes across detection and ranking tasks | Single-Gene Detection, Pairwise Detection, Dominance Ranking, Attribution |

---

## 📥 Audio Access

Audio files are included directly for the `structural_segmentation` config. For all other tasks, each example includes a `youtube_url` field pointing to the source track on YouTube due to copyright restrictions.

---

## 🌍 Multilingual Version (Coming Soon)

We are releasing a multilingual version of BASS covering **9 languages**: Chinese, English, French, Korean, Nigerian Pidgin, Portuguese, Spanish, Vietnamese, and Zulu.

---

## 📬 Contact

For questions or feedback, please open a [GitHub issue](https://github.com/minjang10/bass_music_benchmark/issues) or reach out to us at minjang@cs.washington.edu and oahia@cs.washington.edu.

---

## ✍️ Citation

```bibtex
@article{jang2025bass,
  title   = {BASS: Benchmarking Audio LMs for Musical Structure and Semantic Reasoning},
  author  = {Min Jang and Orevaghene Ahia and Nazif Tamer and Sachin Kumar and Yulia Tsvetkov and Noah A. Smith},
  year    = {2025},
  url     = {https://arxiv.org/abs/2602.04085}
}
```