随着智能合成语音产品的快速发展，人工智能（artificial inteligence, AI）技术在有声书领域的应用日益广泛，为传统的有声读物注入了新的活力。然而，是否 AI有声书的声音越接近真人，就越能受到消费者的青睐？本研究以喜马拉雅平台上的有声书为研究对象，从抓取的音频中提取低维和高维的音频特征，分别建立了随机森林模型和神经网络模型用以预测有声书的声音来源（真人/AI）。基于此，本研究构建了度量声音拟人化程度的指标，并从技术和行为层面分别对该指标的内在机理进行了深入分析，帮助大家理解该拟人化指标如何刻画 AI有声书和真人有声书的异同。最后，本研究通过实证的方法探究了该拟人化指标与消费者行为之间的关系。实证结果表明，有声书的拟人化程度与消费者好感度（即用户评分）之间存在恐怖谷效应，即随着有声书声音的拟人化程度增加，该有声书的评分会呈现先上升后下降的态势。不过，在用户的评论声量方面，恐怖谷效应并不存在。本研究为理解消费者对 AI生成内容的接受程度提供了新的理论和实证视角。

With the rapid advancement of intelligent synthetic voice technology, Al applications in the audiobook industry are injecting new vitality into the field. But does a more human-like Al voice necessarily make audiobooks more appealing to consumers? This study explores this question by analyzing audiobooks on the Himalaya platform. It extracts both low-and high-dimensional audio features from collected audio samples and employs random forest and neural network models to predict whether the audiobook narration originates from a human or Al. Building on these machine learning models, the study constructs an index to quantify the degree of anthropomorphism in audiobook voices. It thoroughly examines the underlying mechanisms of this index from technical and behavioral perspectives, shedding light on how it captures the similarities and diferences between human and Al - generated audiobook narration. Additionally, the paper investigates the relationship between the anthropomorphism index and consumer behavior. Empirical results reveal the presence of an uncanny valley elect: as the degree of anthro-pomorphism increases, audiobook ratings initially rise but subsequently decline. Interestingly, this effect is not observed in the volume of user comments. This research provides both theoretical and empirical insights into consumer acceptance of Al audiobooks, offering a fresh perspective on the evolving dynamics of human—AI interaction in the audiobook market.