Datasets:

Languages:
English
ArXiv:
License:
admin commited on
Commit
4cbd76f
1 Parent(s): f986fe7

upl scripts

Browse files
Files changed (3) hide show
  1. .gitignore +1 -0
  2. README.md +210 -1
  3. pianos.py +258 -0
.gitignore ADDED
@@ -0,0 +1 @@
 
 
1
+ rename.sh
README.md CHANGED
@@ -1,3 +1,212 @@
1
  ---
2
- license: mit
 
 
 
 
 
 
 
 
 
 
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ license: cc-by-nc-nd-4.0
3
+ task_categories:
4
+ - audio-classification
5
+ - image-classification
6
+ language:
7
+ - en
8
+ tags:
9
+ - music
10
+ - art
11
+ pretty_name: Piano Sound Quality Dataset
12
+ size_categories:
13
+ - 10K<n<100K
14
+ viewer: false
15
  ---
16
+
17
+ # Dataset Card for Piano Sound Quality Dataset
18
+ The raw dataset is sourced from the [Piano Sound Quality Dataset](https://ccmusic-database.github.io/en/database/ccm.html#shou1), which includes 12 full-range audio files in .wav/.mp3/.m4a format representing seven models of pianos: Kawai upright piano, Kawai grand piano, Young Change upright piano, Hsinghai upright piano, Grand Theatre Steinway piano, Steinway grand piano, and Pearl River upright piano. Additionally, there are 1,320 split monophonic audio files in .wav/.mp3/.m4a format, bringing the total number of files to 1,332. The dataset also includes a score sheet in .xls format containing subjective evaluations of piano sound quality provided by 29 participants with musical backgrounds.
19
+
20
+ Based on the aforementioned raw dataset, after data processing, we constructed the `default subset` of the current integrated version of the dataset, and its data structure can be viewed in the [viewer](https://www.modelscope.cn/datasets/ccmusic-database/pianos/dataPeview).
21
+
22
+ Due to the need to increase the dataset size and the absence of a popular piano brand, Yamaha, the `default subset` is expanded by recording an upright Yamaha piano into the `8_class subset`.
23
+
24
+ Since the current dataset has been validated by published articles, based on the `8_class subset`, we adopted the data processing method for dataset evaluation from the article and constructed the `eval subset`. Below is a brief introduction to the data structure of each subset.
25
+
26
+ ## Dataset Structure
27
+ ### Default Subset
28
+ <table class="pianos">
29
+ <tr>
30
+ <th>audio</th>
31
+ <th>mel</th>
32
+ <th>label (8-class)</th>
33
+ <th>pitch (88-class)</th>
34
+ </tr>
35
+ <tr>
36
+ <td>.wav, 44100Hz</td>
37
+ <td>.jpg, 44100Hz</td>
38
+ <td>PearlRiver / YoungChang / Steinway-T / Hsinghai / Kawai / Steinway / Kawai-G / Yamaha</td>
39
+ <td>88 pitches on piano</td>
40
+ </tr>
41
+ <tr>
42
+ <td>...</td>
43
+ <td>...</td>
44
+ <td>...</td>
45
+ <td>...</td>
46
+ </tr>
47
+ </table>
48
+
49
+ ### Eval Subset
50
+ <style>
51
+ .pianos td {
52
+ vertical-align: middle !important;
53
+ text-align: center;
54
+ }
55
+ .pianos th {
56
+ text-align: center;
57
+ }
58
+ </style>
59
+ <table class="pianos">
60
+ <tr>
61
+ <th>mel</th>
62
+ <th>label (8-class)</th>
63
+ <th>pitch (88-class)</th>
64
+ </tr>
65
+ <tr>
66
+ <td>.jpg, 0.18s 44100Hz</td>
67
+ <td>PearlRiver / YoungChang / Steinway-T / Hsinghai / Kawai / Steinway / Kawai-G / Yamaha</td>
68
+ <td>88 pitches on piano</td>
69
+ </tr>
70
+ <tr>
71
+ <td>...</td>
72
+ <td>...</td>
73
+ <td>...</td>
74
+ </tr>
75
+ </table>
76
+
77
+ <img src="https://www.modelscope.cn/api/v1/datasets/ccmusic-database/pianos/repo?Revision=master&FilePath=.%2Fdata%2Fpiano.png&View=true">
78
+
79
+ ### Data Instances
80
+ .zip(.wav, jpg)
81
+
82
+ ### Data Fields
83
+ ```
84
+ 1_PearlRiver
85
+ 2_YoungChang
86
+ 3_Steinway-T
87
+ 4_Hsinghai
88
+ 5_Kawai
89
+ 6_Steinway
90
+ 7_Kawai-G
91
+ 8_Yamaha (For Non-default subset)
92
+ ```
93
+
94
+ ### Data Splits for Eval Subset
95
+ | Split | Default | 8_class | Eval |
96
+ | :---------------: | :------------------: | :-----------------: | :-----------------: |
97
+ | train(80%) | 461 | 531 | 14678 |
98
+ | validation(10%) | 59 | 68 | 1835 |
99
+ | test(10%) | 60 | 69 | 1839 |
100
+ | total | 580 | 668 | 18352 |
101
+ | Total duration(s) | `2851.6933333333354` | `3247.941395833335` | `3247.941395833335` |
102
+
103
+ ## Viewer
104
+ <https://www.modelscope.cn/datasets/ccmusic-database/pianos/dataPeview>
105
+
106
+ ## Usage
107
+ ### Default Subset
108
+ ```python
109
+ from datasets import load_dataset
110
+
111
+ ds = load_dataset("ccmusic-database/pianos", name="default")
112
+ for item in ds["train"]:
113
+ print(item)
114
+
115
+ for item in ds["validation"]:
116
+ print(item)
117
+
118
+ for item in ds["test"]:
119
+ print(item)
120
+ ```
121
+
122
+ ### Eval Subset
123
+ ```python
124
+ from datasets import load_dataset
125
+
126
+ ds = load_dataset("ccmusic-database/pianos", name="eval")
127
+ for item in ds["train"]:
128
+ print(item)
129
+
130
+ for item in ds["validation"]:
131
+ print(item)
132
+
133
+ for item in ds["test"]:
134
+ print(item)
135
+ ```
136
+
137
+ ## Maintenance
138
+ ```bash
139
+ GIT_LFS_SKIP_SMUDGE=1 git clone git@hf.co:datasets/ccmusic-database/pianos
140
+ cd pianos
141
+ ```
142
+
143
+ ## Dataset Description
144
+ - **Homepage:** <https://ccmusic-database.github.io>
145
+ - **Repository:** <https://huggingface.co/datasets/CCMUSIC/pianos>
146
+ - **Paper:** <https://doi.org/10.5281/zenodo.5676893>
147
+ - **Leaderboard:** <https://www.modelscope.cn/datasets/ccmusic/pianos>
148
+ - **Point of Contact:** <https://arxiv.org/abs/2310.04722>
149
+
150
+ ### Dataset Summary
151
+ Due to the need to increase the dataset size and the absence of a popular piano brand, Yamaha, the dataset is expanded by recording an upright Yamaha piano in [[1]](https://arxiv.org/pdf/2310.04722.pdf), in which the recording details can be found. This results in a total of 2,020 audio files. As models used in that article require a larger dataset, data augmentation was performed. The original audio was transformed into Mel spectrograms and sliced into 0.18-second segments, a parameter chosen based on empirical experience. This results in 18,745 spectrogram slices. Although 0.18 seconds may seem narrow, this duration is sufficient for the task at hand, as the classification of piano sound quality does not heavily rely on the temporal characteristics of the audio segments.
152
+
153
+ ### Supported Tasks and Leaderboards
154
+ Piano Sound Classification, pitch detection
155
+
156
+ ### Languages
157
+ English
158
+
159
+ ## Dataset Creation
160
+ ### Curation Rationale
161
+ Lack of a dataset for piano sound quality
162
+
163
+ ### Source Data
164
+ #### Initial Data Collection and Normalization
165
+ Zhaorui Liu, Shaohua Ji, Monan Zhou
166
+ #### Who are the source language producers?
167
+ Students from CCMUSIC & CCOM
168
+
169
+ ### Annotations
170
+ #### Annotation process
171
+ Students from CCMUSIC recorded different piano sounds and labeled them, and then a subjective survey of sound quality was conducted to score them.
172
+
173
+ #### Who are the annotators?
174
+ Students from CCMUSIC & CCOM
175
+
176
+ ### Personal and Sensitive Information
177
+ Piano brands
178
+
179
+ ## Considerations for Using the Data
180
+ ### Social Impact of Dataset
181
+ Help develop piano sound quality scoring apps
182
+
183
+ ### Discussion of Biases
184
+ Only for pianos
185
+
186
+ ### Other Known Limitations
187
+ Lack of black keys for Steinway, data imbalance
188
+
189
+ ## Additional Information
190
+ ### Dataset Curators
191
+ Zijin Li
192
+
193
+ ### Evaluation
194
+ [1] [Monan Zhou, Shangda Wu, Shaohua Ji, Zijin Li, and Wei Li. A Holistic Evaluation of Piano Sound Quality[C]//Proceedings of the 10th Conference on Sound and Music Technology (CSMT). Springer, Singapore, 2023.](https://arxiv.org/pdf/2310.04722.pdf)
195
+
196
+ (Note: this paper only uses the first 7 piano classes in the dataset, its future work has finished the 8-class evaluation)
197
+
198
+ ### Citation Information
199
+ ```bibtex
200
+ @dataset{zhaorui_liu_2021_5676893,
201
+ author = {Monan Zhou, Shenyang Xu, Zhaorui Liu, Zhaowen Wang, Feng Yu, Wei Li and Baoqiang Han},
202
+ title = {CCMusic: an Open and Diverse Database for Chinese and General Music Information Retrieval Research},
203
+ month = {mar},
204
+ year = {2024},
205
+ publisher = {HuggingFace},
206
+ version = {1.2},
207
+ url = {https://huggingface.co/ccmusic-database}
208
+ }
209
+ ```
210
+
211
+ ### Contributions
212
+ Provide a dataset for piano sound quality
pianos.py ADDED
@@ -0,0 +1,258 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import os
2
+ import random
3
+ import datasets
4
+ from datasets.tasks import ImageClassification
5
+
6
+
7
+ _NAMES = {
8
+ "PearlRiver": [2.33, 2.53, 2.37, 2.41],
9
+ "YoungChang": [2.53, 2.63, 2.97, 2.71],
10
+ "Steinway-T": [3.6, 3.63, 3.67, 3.63],
11
+ "Hsinghai": [3.4, 3.27, 3.2, 3.29],
12
+ "Kawai": [3.17, 2.5, 2.93, 2.87],
13
+ "Steinway": [4.23, 3.67, 4, 3.97],
14
+ "Kawai-G": [3.37, 2.97, 3.07, 3.14],
15
+ "Yamaha": [3.23, 3.03, 3.17, 3.14],
16
+ }
17
+
18
+ _DBNAME = os.path.basename(__file__).split(".")[0]
19
+
20
+ _HOMEPAGE = f"https://www.modelscope.cn/datasets/ccmusic-database/{_DBNAME}"
21
+
22
+ _DOMAIN = f"https://www.modelscope.cn/api/v1/datasets/ccmusic-database/{_DBNAME}/repo?Revision=master&FilePath=data"
23
+
24
+
25
+ _PITCHES = {
26
+ "009": "A2",
27
+ "010": "A2#/B2b",
28
+ "011": "B2",
29
+ "100": "C1",
30
+ "101": "C1#/D1b",
31
+ "102": "D1",
32
+ "103": "D1#/E1b",
33
+ "104": "E1",
34
+ "105": "F1",
35
+ "106": "F1#/G1b",
36
+ "107": "G1",
37
+ "108": "G1#/A1b",
38
+ "109": "A1",
39
+ "110": "A1#/B1b",
40
+ "111": "B1",
41
+ "200": "C",
42
+ "201": "C#/Db",
43
+ "202": "D",
44
+ "203": "D#/Eb",
45
+ "204": "E",
46
+ "205": "F",
47
+ "206": "F#/Gb",
48
+ "207": "G",
49
+ "208": "G#/Ab",
50
+ "209": "A",
51
+ "210": "A#/Bb",
52
+ "211": "B",
53
+ "300": "c",
54
+ "301": "c#/db",
55
+ "302": "d",
56
+ "303": "d#/eb",
57
+ "304": "e",
58
+ "305": "f",
59
+ "306": "f#/gb",
60
+ "307": "g",
61
+ "308": "g#/ab",
62
+ "309": "a",
63
+ "310": "a#/bb",
64
+ "311": "b",
65
+ "400": "c1",
66
+ "401": "c1#/d1b",
67
+ "402": "d1",
68
+ "403": "d1#/e1b",
69
+ "404": "e1",
70
+ "405": "f1",
71
+ "406": "f1#/g1b",
72
+ "407": "g1",
73
+ "408": "g1#/a1b",
74
+ "409": "a1",
75
+ "410": "a1#/b1b",
76
+ "411": "b1",
77
+ "500": "c2",
78
+ "501": "c2#/d2b",
79
+ "502": "d2",
80
+ "503": "d2#/e2b",
81
+ "504": "e2",
82
+ "505": "f2",
83
+ "506": "f2#/g2b",
84
+ "507": "g2",
85
+ "508": "g2#/a2b",
86
+ "509": "a2",
87
+ "510": "a2#/b2b",
88
+ "511": "b2",
89
+ "600": "c3",
90
+ "601": "c3#/d3b",
91
+ "602": "d3",
92
+ "603": "d3#/e3b",
93
+ "604": "e3",
94
+ "605": "f3",
95
+ "606": "f3#/g3b",
96
+ "607": "g3",
97
+ "608": "g3#/a3b",
98
+ "609": "a3",
99
+ "610": "a3#/b3b",
100
+ "611": "b3",
101
+ "700": "c4",
102
+ "701": "c4#/d4b",
103
+ "702": "d4",
104
+ "703": "d4#/e4b",
105
+ "704": "e4",
106
+ "705": "f4",
107
+ "706": "f4#/g4b",
108
+ "707": "g4",
109
+ "708": "g4#/a4b",
110
+ "709": "a4",
111
+ "710": "a4#/b4b",
112
+ "711": "b4",
113
+ "800": "c5",
114
+ }
115
+
116
+ _URLS = {
117
+ "audio": f"{_DOMAIN}/audio.zip",
118
+ "mel": f"{_DOMAIN}/mel.zip",
119
+ "eval": f"{_DOMAIN}/eval.zip",
120
+ }
121
+
122
+
123
+ class pianos(datasets.GeneratorBasedBuilder):
124
+ def _info(self):
125
+ names = list(_NAMES.keys())
126
+ if self.config.name == "default":
127
+ names = names[:-1]
128
+
129
+ return datasets.DatasetInfo(
130
+ features=(
131
+ datasets.Features(
132
+ {
133
+ "audio": datasets.Audio(sampling_rate=44100),
134
+ "mel": datasets.Image(),
135
+ "label": datasets.features.ClassLabel(names=names),
136
+ "pitch": datasets.features.ClassLabel(
137
+ names=list(_PITCHES.values())
138
+ ),
139
+ "bass_score": datasets.Value("float32"),
140
+ "mid_score": datasets.Value("float32"),
141
+ "treble_score": datasets.Value("float32"),
142
+ "avg_score": datasets.Value("float32"),
143
+ }
144
+ )
145
+ if self.config.name != "eval"
146
+ else datasets.Features(
147
+ {
148
+ "mel": datasets.Image(),
149
+ "label": datasets.features.ClassLabel(names=names),
150
+ "pitch": datasets.features.ClassLabel(
151
+ names=list(_PITCHES.values())
152
+ ),
153
+ "bass_score": datasets.Value("float32"),
154
+ "mid_score": datasets.Value("float32"),
155
+ "treble_score": datasets.Value("float32"),
156
+ "avg_score": datasets.Value("float32"),
157
+ }
158
+ )
159
+ ),
160
+ homepage=_HOMEPAGE,
161
+ license="CC-BY-NC-ND",
162
+ version="1.2.0",
163
+ supervised_keys=("mel", "label"),
164
+ task_templates=ImageClassification(
165
+ image_column="mel",
166
+ label_column="label",
167
+ ),
168
+ )
169
+
170
+ def _split_generators(self, dl_manager):
171
+ dataset = []
172
+ if self.config.name != "eval":
173
+ subset = {}
174
+ audio_files = dl_manager.download_and_extract(_URLS["audio"])
175
+ for path in dl_manager.iter_files([audio_files]):
176
+ fname = os.path.basename(path)
177
+ if fname.endswith(".wav"):
178
+ lebal = os.path.basename(os.path.dirname(path))
179
+ if self.config.name == "default" and lebal == "Yamaha":
180
+ continue
181
+
182
+ subset[fname.split(".")[0]] = {
183
+ "audio": path,
184
+ "label": lebal,
185
+ "pitch": _PITCHES[fname[1:4]],
186
+ "bass_score": _NAMES[lebal][0],
187
+ "mid_score": _NAMES[lebal][1],
188
+ "treble_score": _NAMES[lebal][2],
189
+ "avg_score": _NAMES[lebal][3],
190
+ }
191
+
192
+ mel_files = dl_manager.download_and_extract(_URLS["mel"])
193
+ for path in dl_manager.iter_files([mel_files]):
194
+ fname = os.path.basename(path)
195
+ pname = fname.split(".")[0]
196
+ if fname.endswith(".jpg") and pname in subset:
197
+ subset[pname]["mel"] = path
198
+
199
+ dataset = list(subset.values())
200
+
201
+ else:
202
+ data_files = dl_manager.download_and_extract(_URLS["eval"])
203
+ for path in dl_manager.iter_files([data_files]):
204
+ fname: str = os.path.basename(path)
205
+ if fname.endswith(".jpg"):
206
+ lebal = os.path.basename(os.path.dirname(path))
207
+ dataset.append(
208
+ {
209
+ "mel": path,
210
+ "label": lebal,
211
+ "pitch": _PITCHES[fname.split("_")[0]],
212
+ "bass_score": _NAMES[lebal][0],
213
+ "mid_score": _NAMES[lebal][1],
214
+ "treble_score": _NAMES[lebal][2],
215
+ "avg_score": _NAMES[lebal][3],
216
+ }
217
+ )
218
+
219
+ names = list(_NAMES.keys())
220
+ if self.config.name == "default":
221
+ names = names[:-1]
222
+
223
+ categories = {}
224
+ for name in names:
225
+ categories[name] = []
226
+
227
+ for data in dataset:
228
+ categories[data["label"]].append(data)
229
+
230
+ testset, validset, trainset = [], [], []
231
+ for cls in categories:
232
+ random.shuffle(categories[cls])
233
+ count = len(categories[cls])
234
+ p80 = int(count * 0.8)
235
+ p90 = int(count * 0.9)
236
+ trainset += categories[cls][:p80]
237
+ validset += categories[cls][p80:p90]
238
+ testset += categories[cls][p90:]
239
+
240
+ random.shuffle(trainset)
241
+ random.shuffle(validset)
242
+ random.shuffle(testset)
243
+
244
+ return [
245
+ datasets.SplitGenerator(
246
+ name=datasets.Split.TRAIN, gen_kwargs={"files": trainset}
247
+ ),
248
+ datasets.SplitGenerator(
249
+ name=datasets.Split.VALIDATION, gen_kwargs={"files": validset}
250
+ ),
251
+ datasets.SplitGenerator(
252
+ name=datasets.Split.TEST, gen_kwargs={"files": testset}
253
+ ),
254
+ ]
255
+
256
+ def _generate_examples(self, files):
257
+ for i, path in enumerate(files):
258
+ yield i, path