Trust-UBA: Detecting trust from the trustor's voice

Abstract

The protocol of the dataset consists of an interactive session in which a subject answers a series of factual questions with the help of a virtual assistant (VA). Before each series, subjects are told that the VA was previously rated by other users as either competent or incompetent; the VA then answers the questions consistently with its alleged ability, answering some of them wrongly in the incompetent condition. Each subject solved two series of 18 questions, one in each condition. All interactions are speech-based, with subjects and VA communicating verbally, which allows the recording of speech produced under different trust conditions. The dataset comprises the speech recordings of the subjects' queries to the VA, the answers given to each question, and the responses to the personal information, familiarity with and trust in technology, personality, intermediate and final trust questionnaires, for two versions of the experiment: V1, with 84 subjects (50 in the laboratory and 34 remote), which also includes external annotations of trust and emotions by expert annotators, and V2, with 62 remote subjects, which also includes neutral speech recordings (free speech and reading). The speech was collected from speakers of Argentine Spanish.