Chills experienced in response to music listening have been linked to both happiness and sadness expressed by music. To investigate these conflicting effects of valence on chills, we conducted a computational analysis on a corpus of 988 tracks previously reported to elicit chills, by comparing them with a control set of tracks matched by artist, duration, and popularity. We analysed track-level audio features obtained with the Spotify Web API across the two sets of tracks, resulting in confirmatory findings that tracks which cause chills were sadder than matched tracks and exploratory findings that they were also slower, less intense, and more instrumental than matched tracks on average. We also found that the audio characteristics of chills tracks were related to the direction and magnitude of the difference in valence between the two sets of tracks. We discuss these results in light of the current literature on valence and chills in music, provide a new interpretation in terms of personality correlates of musical preference, and review the advantages and limitations of our computational approach.

听音乐时产生的寒意与音乐表达的快乐和悲伤有关。为了研究效价对寒意的这些相互矛盾的影响，我们对先前报道的会引起寒意的 988 首曲目的语料库进行了计算分析，将它们与与艺术家、持续时间和流行度相匹配的对照曲目集进行比较。我们分析了通过 Spotify Web API 在两组曲目中获得的曲目级音频特征，得出了确认性的发现，即引起寒意的曲目比匹配的曲目更悲伤，而探索性的发现则表明，它们也比匹配的曲目更慢、不那么激烈且更有帮助性。平均匹配曲目。我们还发现，chill 曲目的音频特征与两组曲目之间价差的方向和大小有关。我们根据当前有关音乐效价和寒意的文献讨论了这些结果，就音乐偏好的个性相关性提供了新的解释，并回顾了我们的计算方法的优点和局限性。