File size: 3,753 Bytes
4021603
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
---
language: 
- en
pretty_name: MoisesDB
tags:
- audio
- music
- source separation
license: other
license_name: cc-by-nc-sa-4.0
license_link: https://creativecommons.org/licenses/by-nc-sa/4.0/
---

# MoisesDB
Moises Dataset for Source Separation

### Dataset Description

- **Homepage:** [MoisesDB homepage](https://developer.moises.ai/research/)
- **Repository:** [MoisesDB repository](https://github.com/moises-ai/moises-db)
- **Paper:** [Moisesdb: A dataset for source separation beyond 4-stems](https://arxiv.org/abs/2307.15913)
- **Point of Contact:** [Igor Pereira](mailto:igor@moises.ai)

### Dataset Summary

MoisesDB is a dataset for source separation. It provides a collection of tracks and their separated stems (vocals, bass, drums, etc.). The dataset is used to evaluate the performance of source separation algorithms.

# Download the data

Please download the dataset at our research [website](https://developer.moises.ai/research), extract it and configure the environment variable `MOISESDB_PATH` accordingly.

```
export MOISESDB_PATH=./moises-db-data
```

The directory structure should be

```
moisesdb:
    moisesdb_v0.1
        track uuid 0
        track uuid 1
        .
        .
        .
```

# Install

You can install this package with

```
pip install git+https://github.com/moises-ai/moises-db.git
```

# Usage

## `MoisesDB`

After downloading and configuring the path for the dataset, you can create an instance of `MoisesDB` to access the tracks. You can also provide the dataset path with the `data_path` argument.

```
from moisesdb.dataset import MoisesDB

db = MoisesDB(
    data_path='./moisesdb',
    sample_rate=44100
)
```

The `MoisesDB` object has iterator properties that you can use to access all files within the dataset.

```
n_songs = len(db)
track = db[0]  # Returns a MoisesDBTrack object
```

## `MoisesDBTrack`

The `MoisesDBTrack` object holds information about a track in the dataset, perform on-the-fly mixing for stems and multiple sources within a stem.

You can access all the stems and mixture from the `stem` and `audio` properties. The `stem` property returns a dictionary whith available stems as keys and `nd.array` on values. The `audio` property results in a `nd.array` with the mixture.

```
track = db[0]
stems = track.stems  # stems = {'vocals': ..., 'bass': ..., ...}
mixture track.audio # mixture = nd.array
```

The `MoisesDBTrack` object also contains other non-audio information from the track such as:
- `track.id`
- `track.provider`
- `track.artist`
- `track.name`
- `track.genre`
- `track.sources`
- `track.bleedings`
- `track.activity`

The stems and mixture are computed on-the-fly. You can create a stems-only version of the dataset using the `save_stems` method of the `MoisesDBTrack`.

```
track = db[0]
path =  './moises-db-stems/0'
track.save_stems(path)
```

# Performance Evaluation

We run a few source separation algorithms as well as oracle methods to evaluate the performance of each track of the `MoisesDB`. These results are located in `csv` files at the `benchmark` folder.

# Citing

If you used the `MoisesDB` dataset on your research, please cite the following paper.

```
@misc{pereira2023moisesdb,
      title={Moisesdb: A dataset for source separation beyond 4-stems}, 
      author={Igor Pereira and Felipe Araújo and Filip Korzeniowski and Richard Vogl},
      year={2023},
      eprint={2307.15913},
      archivePrefix={arXiv},
      primaryClass={cs.SD}
}
```

# Licensing

`MoisesDB` is distributed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License (CC BY-NC-SA 4.0).

For the complete license terms, please visit: https://creativecommons.org/licenses/by-nc-sa/4.0/

See [LICENSE](LICENSE) file for details.