Update README.md
Browse filesInitial README.md
README.md
CHANGED
@@ -1,27 +1,108 @@
|
|
1 |
-
---
|
2 |
-
license: mit
|
3 |
-
dataset_info:
|
4 |
-
features:
|
5 |
-
- name: track_name
|
6 |
-
dtype: string
|
7 |
-
- name: start_time
|
8 |
-
dtype: int64
|
9 |
-
- name: embedding
|
10 |
-
dtype:
|
11 |
-
array2_d:
|
12 |
-
shape:
|
13 |
-
- 240
|
14 |
-
- 4800
|
15 |
-
dtype: float32
|
16 |
-
splits:
|
17 |
-
- name: train
|
18 |
-
num_bytes: 4166537771
|
19 |
-
num_examples: 904
|
20 |
-
download_size: 4171864391
|
21 |
-
dataset_size: 4166537771
|
22 |
-
configs:
|
23 |
-
- config_name: default
|
24 |
-
data_files:
|
25 |
-
- split: train
|
26 |
-
path: data/train-*
|
27 |
-
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: mit
|
3 |
+
dataset_info:
|
4 |
+
features:
|
5 |
+
- name: track_name
|
6 |
+
dtype: string
|
7 |
+
- name: start_time
|
8 |
+
dtype: int64
|
9 |
+
- name: embedding
|
10 |
+
dtype:
|
11 |
+
array2_d:
|
12 |
+
shape:
|
13 |
+
- 240
|
14 |
+
- 4800
|
15 |
+
dtype: float32
|
16 |
+
splits:
|
17 |
+
- name: train
|
18 |
+
num_bytes: 4166537771
|
19 |
+
num_examples: 904
|
20 |
+
download_size: 4171864391
|
21 |
+
dataset_size: 4166537771
|
22 |
+
configs:
|
23 |
+
- config_name: default
|
24 |
+
data_files:
|
25 |
+
- split: train
|
26 |
+
path: data/train-*
|
27 |
+
---
|
28 |
+
|
29 |
+
# Jukebox Embeddings for the URMP Dataset
|
30 |
+
|
31 |
+
## Overview
|
32 |
+
|
33 |
+
This dataset extends the University of Rochester Multi-Modal Music Performance (URMP) Dataset by providing embeddings for each audio file.
|
34 |
+
|
35 |
+
## Original URMP Dataset
|
36 |
+
|
37 |
+
[Link to official site](https://labsites.rochester.edu/air/projects/URMP.html)
|
38 |
+
|
39 |
+
The URMP dataset was created to facilitate audio-visual analysis of musical performances. It comprises multiple simple multi-instrument musical pieces assembled from coordinated but separately recorded performances of individual tracks.
|
40 |
+
|
41 |
+
## Jukebox Embeddings
|
42 |
+
|
43 |
+
Embeddings are derived from [OpenAI's Jukebox model](https://openai.com/index/jukebox/), following the approach described in [Castellon et al. (2021)](https://arxiv.org/abs/2107.05677) with some modifications followed in [Spotify's Llark paper](https://arxiv.org/pdf/2310.07160):
|
44 |
+
|
45 |
+
- Source: Output of the 36th layer of the Jukebox encoder
|
46 |
+
- Original Jukebox encoding: 4800-dimensional vectors at 345Hz
|
47 |
+
- Audio/embeddings are chunked into 25 seconds clips as that is the max Jukebox can take in as input, any clips shorter than 25 seconds are padded before passed through Jukebox
|
48 |
+
- Approach: Mean-pooling within 100ms frames, resulting in:
|
49 |
+
- Downsampled frequency: 10Hz
|
50 |
+
- Embedding size: 1.2 × 10^6 for a 25s audio clip
|
51 |
+
- Embedding shape: [240, 4800]
|
52 |
+
- This method retains temporal information while reducing the embedding size
|
53 |
+
|
54 |
+
### Why Jukebox? Are these embeddings state-of-the-art as of September 2024?
|
55 |
+
|
56 |
+
Determining the optimal location to extract embeddings from large models typically requires extensive probing. This involves testing various activations or extracted layers of the model on different classification tasks through a process of trial and error. Additional fine-tuning is often done to optimise embeddings across these tasks.
|
57 |
+
|
58 |
+
The two largest publicly available music generation and music continuation (i.e.: able to take in audio as input) models are Jukebox and MusicGen. According to [this paper on probing MusicGen](https://www.merl.com/publications/docs/TR2024-032.pdf), embeddings extracted from Jukebox appears to outperform MusicGen on average in their classification tasks.
|
59 |
+
|
60 |
+
## Dataset Features
|
61 |
+
|
62 |
+
This extension to the URMP dataset includes:
|
63 |
+
|
64 |
+
1. File name of each WAV file in the URMP dataset
|
65 |
+
2. Start time of the audio
|
66 |
+
3. Jukebox embedding for each audio file
|
67 |
+
|
68 |
+
## Applications
|
69 |
+
|
70 |
+
This extended dataset can be used for various tasks, including but not limited to:
|
71 |
+
|
72 |
+
- Music source separation
|
73 |
+
- Transcription
|
74 |
+
- Performance analysis
|
75 |
+
- Multi-modal information retrieval
|
76 |
+
|
77 |
+
## Usage
|
78 |
+
|
79 |
+
```python
|
80 |
+
from datasets import load_dataset
|
81 |
+
|
82 |
+
dataset = load_dataset("jonflynn/urmp_jukebox_embeddings")
|
83 |
+
|
84 |
+
# There's only one split, that is train
|
85 |
+
train_dataset = dataset['train']
|
86 |
+
```
|
87 |
+
|
88 |
+
## Citation
|
89 |
+
|
90 |
+
If you use this dataset in your research, please cite the original URMP paper and this extension:
|
91 |
+
|
92 |
+
```bibtex
|
93 |
+
@article{li2018creating,
|
94 |
+
title={Creating a multi-track classical music performance dataset for multi-modal music analysis: Challenges, insights, and applications},
|
95 |
+
author={Li, Bochen and Liu, Xinzhao and Dinesh, Karthik and Duan, Zhiyao and Sharma, Gaurav},
|
96 |
+
journal={IEEE Transactions on Multimedia},
|
97 |
+
year={2018},
|
98 |
+
publisher={IEEE}
|
99 |
+
}
|
100 |
+
|
101 |
+
@dataset{flynn2024urmpjukebox,
|
102 |
+
author = {Jon Flynn},
|
103 |
+
title = {Jukebox Embeddings for the URMP Dataset},
|
104 |
+
year = {2024},
|
105 |
+
publisher = {Hugging Face},
|
106 |
+
howpublished = {\url{https://huggingface.co/datasets/jonflynn/urmp_jukebox_embeddings}},
|
107 |
+
}
|
108 |
+
```
|