---
language:
- en
license: cc-by-nc-sa-4.0
task_categories:
- audio-to-audio
pretty_name: A2LS test-100
---

# A2LS test-100

A 100-sample subset (random seed 42) of the test split of the Audio-to-Lead-Sheet
(A2LS) dataset, with pre-computed [MuQ](https://github.com/tencent-ailab/MuQ)
features. The audio itself is **not** included.

Each row of the `test` split contains:

| column        | content                                                                            |
| ------------- | ---------------------------------------------------------------------------------- |
| `file_name`   | SheetSage record ID                                                                |
| `features`    | pre-computed MuQ features                                                          |
| `beat_onsets` | beat onset times in seconds                                                        |
| `length`      | number of feature frames                                                           |
| `transcript`  | reference token sequence                                                           |
| `kern`        | ground-truth lead sheet in Humdrum `**kern` (**kern**, **cdata**, **text** spines) |

Load with:

```python
from datasets import load_from_disk
dataset = load_from_disk("data/a2ls-100")["test"]
```

Data is released under CC-BY-NC-SA 4.0.