CAL500 (Computer Audition Lab 500)

CAL500（Computer Audition Lab 500）是一个旨在评估音乐信息检索系统的数据集。它由来自西方流行音乐的 502 首歌曲组成。音频表示为前 13 个 Mel 频率倒谱系数（及其一阶和二阶导数）的时间序列，通过在每首歌曲的波形上滑动 12 ms 半重叠短时间窗口来提取。每首歌曲都由至少 3 个人注释，包含 135 个音乐相关概念，涵盖 6 个语义类别：29 个乐器被注释为歌曲中是否存在，22 个声乐特征被注释为与歌手是否相关，36 个流派，18 个情绪以 1 到 3 的等级进行评分（例如，不开心“、中性”、“开心”），15 个歌曲概念描述了歌曲、艺术家和录音的音质（例如，节奏、能量、音质） , 15 个使用术语（例如，“我会在开车、睡觉等时听这首歌”）。

CAL500 (Computer Audition Lab 500) is a dataset designed for evaluating music information retrieval systems. It consists of 502 tracks from Western popular music. The audio data is represented as time series of the first 13 Mel-frequency cepstral coefficients (and their first- and second-order derivatives), extracted by sliding a 12-ms half-overlapping short-time window across the waveform of each track. Each track is annotated by at least three annotators, covering 135 music-related concepts categorized into six semantic classes: 29 instruments annotated for their presence in the track; 22 vocal characteristics annotated in relation to the singer’s presence; 36 music genres; 18 emotions rated on a scale of 1 to 3 (e.g., "sad", "neutral", "happy"); 15 track concepts describing the timbral qualities of the track, artist and recording (e.g., tempo, energy, timbre); and 15 usage terms (e.g., "I would listen to this track while driving, sleeping, etc.")

搜集汇总

背景与挑战

背景概述

CAL500是一个用于音乐信息检索评估的数据集，包含502首西方流行歌曲，音频以Mel频率倒谱系数时间序列表示。数据集特点在于每首歌曲由多人注释，覆盖135个音乐概念，包括乐器、情绪、流派等6个类别，适用于多标签分类和语义分析任务。