Dataset Card for Acapella Evaluation

The original dataset, sourced from the Acapella Evaluation Dataset, comprises six Mandarin pop song segments performed by 22 singers, resulting in a total of 132 audio clips. Each segment includes both a verse and a chorus. Four judges from the China Conservatory of Music assess the singing across nine dimensions: pitch, rhythm, vocal range, timbre, pronunciation, vibrato, dynamics, breath control, and overall performance, using a 10-point scale. The evaluations are recorded in an Excel spreadsheet in .xls format.

Due to the original dataset comprising separate files for audio recordings and evaluation sheets, which hindered efficient data retrieval, we combined the original vocal recordings with their corresponding evaluation sheets to construct the default subset of the current integrated version of the dataset. The data structure can be viewed in the viewer. The current dataset is already endorsed by published articles, hence there is no need to construct the eval subset.

Viewer

https://www.modelscope.cn/datasets/ccmusic-database/acapella/dataPeview

Dataset Structure

audio	mel	singer_id	pitch / rhythm / ... / overall_performance (9 colums)
.wav, 48000Hz	.jpg, 48000Hz	int	float(0-10)
...	...	...	...

Data Instances

.zip(.wav), .csv

Data Fields

song, singer id, pitch, rhythm, vocal range, timbre, pronunciation, vibrato, dynamic, breath control and overall performance

Data Splits

song1-6

Dataset Summary

Due to the original dataset comprising separate files for audio recordings and evaluation sheets, which hindered efficient data retrieval, we have consolidated the raw vocal recordings with their corresponding assessments. The dataset is divided into six segments, each representing a different song, resulting in a total of six divisions. Each segment contains 22 entries, with each entry detailing the vocal recording of an individual singer sampled at 22,050 Hz, the singer's ID, and evaluations across the nine dimensions previously mentioned. Consequently, each entry encompasses 11 columns of data. This dataset is well-suited for tasks such as vocal analysis and regression-based singing voice rating. For instance, as previously stated, the final column of each entry denotes the overall performance score, allowing the audio to be utilized as data and this score to serve as the label for regression analysis.

Supported Tasks and Leaderboards

Acapella evaluation/scoring

Languages

Chinese, English

Maintenance

git clone git@hf.co:datasets/ccmusic-database/acapella
cd acapella

Usage

from datasets import load_dataset

dataset = load_dataset("ccmusic-database/acapella", subset="default")
for i in range(1, 7):
    for item in dataset[f"song{i}"]:
        print(item)

Dataset Creation

Curation Rationale

Lack of a training dataset for the acapella scoring system

Source Data

Initial Data Collection and Normalization

Zhaorui Liu, Monan Zhou

Who are the source language producers?

Students and judges from CCMUSIC

Annotations

Annotation process

6 Mandarin song segments were sung by 22 singers, totaling 132 audio clips. Each segment consists of a verse and a chorus. Four judges evaluate the singing from nine aspects which are pitch, rhythm, vocal range, timbre, pronunciation, vibrato, dynamic, breath control and overall performance on a 10-point scale. The scores are recorded on a sheet.

Who are the annotators?

Judges from CCMUSIC

Personal and Sensitive Information

Singers' and judges' names are hided

Considerations for Using the Data

Social Impact of Dataset

Providing a training dataset for the acapella scoring system may improve the development of related Apps

Discussion of Biases

Only for Mandarin songs

Other Known Limitations

No starting point has been marked for the vocal

Additional Information

Dataset Curators

Zijin Li

Evaluation

Li, R.; Zhang, M. Singing-Voice Timbre Evaluations Based on Transfer Learning. Appl. Sci. 2022, 12, 9931. https://doi.org/10.3390/app12199931

Citation Information

@dataset{zhaorui_liu_2021_5676893,
  author       = {Monan Zhou, Shenyang Xu, Zhaorui Liu, Zhaowen Wang, Feng Yu, Wei Li and Baoqiang Han},
  title        = {CCMusic: an Open and Diverse Database for Chinese and General Music Information Retrieval Research},
  month        = {mar},
  year         = {2024},
  publisher    = {HuggingFace},
  version      = {1.2},
  url          = {https://huggingface.co/ccmusic-database}
}

Contributions

Provide a training dataset for the acapella scoring system