Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,85 @@
|
|
1 |
---
|
2 |
license: mit
|
|
|
|
|
|
|
|
|
|
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: mit
|
3 |
+
tags:
|
4 |
+
- music
|
5 |
+
pretty_name: FMA rank
|
6 |
+
size_categories:
|
7 |
+
- 100K<n<1M
|
8 |
---
|
9 |
+
|
10 |
+
# What is FMA-rank?
|
11 |
+
|
12 |
+
FMA is a music dataset from the Free Music Archive, containing over 8000 hours of Creative Commons-licensed music from 107k tracks across 16k artists and 15k albums.
|
13 |
+
It was created in 2017 by [Defferrard et al.](https://arxiv.org/abs/1612.01840) in collaboration with [Free Music Archive](https://freemusicarchive.org/).
|
14 |
+
|
15 |
+
FMA contains a lot of good music, and a lot of bad music, so the question is: can we rank the samples in FMA?
|
16 |
+
|
17 |
+
FMA-rank is a CLAP-based statistical ranking of each sample in FMA. We calculate the log-likelihood of each sample in FMA belonging to an estimated gaussian in the CLAP latent space, using these values we can rank and filter FMA. In log-likelihood, higher values are better.
|
18 |
+
|
19 |
+
# Quickstart
|
20 |
+
|
21 |
+
Download any FMA split from the official github https://github.com/mdeff/fma. Extract the FMA folder from the downloaded zip and set the path to the folder in `fma_root_dir`.
|
22 |
+
Run the following code snippet to load and filter the FMA samples according to the given percentages. The code snippet will return a HF audio dataset.
|
23 |
+
```Python
|
24 |
+
from datasets import load_dataset, Dataset, Audio
|
25 |
+
import os
|
26 |
+
|
27 |
+
# provide location of fma folder
|
28 |
+
fma_root_dir = "/path/to/fma/folder"
|
29 |
+
|
30 |
+
# provide percentage of fma dataset to use
|
31 |
+
# for whole dataset, use start_percentage=0 and end_percentage=100
|
32 |
+
# for worst 20% of dataset, use start_percentage=0 and end_percentage=20
|
33 |
+
# for best 20% of dataset, use the following values:
|
34 |
+
start_percentage = 80
|
35 |
+
end_percentage = 100
|
36 |
+
|
37 |
+
# load fma_rank.csv from huggingface and sort from lowest to highest
|
38 |
+
csv_loaded = load_dataset("DISCOX/FMA-rank")
|
39 |
+
fma_item_list = csv_loaded["train"]
|
40 |
+
fma_sorted_list = sorted(fma_item_list, key=lambda d: d['CLAP-log-likelihood'])
|
41 |
+
|
42 |
+
def parse_fma_audio_folder(fma_root_dir):
|
43 |
+
valid_fma_ids = []
|
44 |
+
subfolders = os.listdir(fma_root_dir)
|
45 |
+
for subfolder in subfolders:
|
46 |
+
subfolder_path = os.path.join(fma_root_dir, subfolder)
|
47 |
+
if os.path.isdir(subfolder_path):
|
48 |
+
music_files = os.listdir(subfolder_path)
|
49 |
+
for music_file in music_files:
|
50 |
+
if ".mp3" not in music_file:
|
51 |
+
continue
|
52 |
+
else:
|
53 |
+
fma_id = music_file.split('.')[0]
|
54 |
+
valid_fma_ids.append(fma_id)
|
55 |
+
return valid_fma_ids
|
56 |
+
|
57 |
+
# select the existing files according to the provided fma folder
|
58 |
+
valid_fma_ids = parse_fma_audio_folder(fma_root_dir)
|
59 |
+
df_dict = {"id":[], "score": [], "audio": []}
|
60 |
+
for fma_item in fma_sorted_list:
|
61 |
+
this_id = f"{fma_item['id']:06d}"
|
62 |
+
if this_id in valid_fma_ids:
|
63 |
+
df_dict["id"].append(this_id)
|
64 |
+
df_dict["score"].append(fma_item["CLAP-log-likelihood"])
|
65 |
+
df_dict["audio"].append(os.path.join(fma_root_dir, this_id[:3] , this_id+".mp3"))
|
66 |
+
|
67 |
+
# filter the fma dataset according to the percentage defined above
|
68 |
+
i_start = int(start_percentage * len(df_dict["id"]) / 100)
|
69 |
+
i_end = int(end_percentage * len(df_dict["id"]) / 100)
|
70 |
+
df_dict_filtered = {
|
71 |
+
"id": df_dict["id"][i_start:i_end],
|
72 |
+
"score": df_dict["score"][i_start:i_end],
|
73 |
+
"audio": df_dict["audio"][i_start:i_end],
|
74 |
+
}
|
75 |
+
|
76 |
+
# get final dataset
|
77 |
+
audio_dataset = Dataset.from_dict(df_dict_filtered).cast_column("audio", Audio())
|
78 |
+
|
79 |
+
"""
|
80 |
+
Dataset({
|
81 |
+
features: ['id', 'score', 'audio'],
|
82 |
+
num_rows: 1599
|
83 |
+
})
|
84 |
+
"""
|
85 |
+
```
|