Synthia’s melody
收藏arXiv2023-09-26 更新2024-06-21 收录
下载链接：
https://github.com/cynthpie/Synthia_melody
下载链接
链接失效反馈官方服务：
资源简介：
Synthia’s melody是由帝国理工学院伦敦分院和奥格斯堡大学合作开发的一个音频数据生成框架，旨在模拟无限多样性的4秒旋律，具有用户指定的音乐键、音色和响度等特征。该数据集通过预设脚本生成，确保实验结果的可重复性和可比较性。数据集包含50000条4秒的旋律样本，格式为.wav，采样率为16000 Hz。Synthia’s melody特别适用于评估音频深度学习模型在不同数据分布偏移下的表现，如领域偏移和样本选择偏差，为音频领域的无监督域适应研究提供了一个强大的测试平台。
Synthia’s melody is an audio data generation framework co-developed by Imperial College London and the University of Augsburg. It aims to generate infinitely diverse 4-second melodies with user-specified musical characteristics including key, timbre, and loudness. This dataset is generated via preset scripts, ensuring the reproducibility and comparability of experimental results. It contains 50,000 4-second melody samples in WAV format with a sampling rate of 16000 Hz. Synthia’s melody is particularly suitable for evaluating the performance of audio deep learning models under various data distribution shifts such as domain shift and sample selection bias, providing a powerful benchmark platform for unsupervised domain adaptation research in the audio domain.
提供机构：
帝国理工学院伦敦分院，英国 †奥格斯堡大学，德国
创建时间：
2023-09-26
