真假声数据集

原始数据集包含1280个用真声/假声演唱的单音音频文件（.wav格式），真声被标记为_chest，假声被标记为_falsetto。此外还收录了每段音频的标注标签、mel spectrogram、MFCC、频谱特征，共计5120个.csv文件。经处理后的数据集是由1280个单音音频文件转成的频谱切片构成的，共包含男真声、男假声、女真声、女假声4个类别，数据量总计8974

The original dataset contains 1280 monophonic audio files in .wav format, sung with chest voice and falsetto. Chest voice is labeled as _chest, while falsetto is labeled as _falsetto. Additionally, it encompasses annotation labels, mel spectrograms, MFCCs, and spectral features corresponding to each audio clip, with a total of 5120 .csv files. The processed dataset is constructed from spectral slices converted from the 1280 monophonic audio files, which includes four categories: male chest voice, male falsetto, female chest voice, and female falsetto, with a total of 8974 samples.

搜集汇总

背景与挑战

背景概述

该数据集包含1,280个单音演唱音频文件，由中国音乐学院声乐专业学生录制，真声和假声分别标记为'chest'和'falsetto'，并提供了Mel频谱图、MFCC等特征文件。原始数据未区分性别，后经手动审核添加了性别注释，构建了默认子集和校验子集，用于音频分类任务，如歌唱性别和真假声分类。