Towards Modelling Trust in Voice at Zero Acquaintance

Abstract

Trust is essential in many human relationships, especially where there is an element of inter-dependency. However, humans tend to make quick judgements about trusting other individuals, even those met at zero acquaintance. Past studies have shown the significance of voice in perceived trustworthiness, but research associating trustworthiness and different vocal features such as speech rate and fundamental frequency (f0) has yet to yield consistent results. Therefore, this paper proposes a method to investigate 1) the association between trustworthiness and different vocal features, 2) the vocal characteristics that Malaysian ethnic groups base their judgement of trustworthiness on and 3) building a neural network model that predicts the degree of trustworthiness in a human voice. In the method proposed, a reliable set of audio clips will be obtained and analyzed with SoundGen to determine the acoustical characteristics. Then the audio clips will be distributed to a large group of untrained respondents to rate their degree of trust in the speakers of each audio clip. The participants will be able to choose from 30 sets of audio clips which will consist of 6 audio clips each. The acoustic characteristics will be analyzed and com-pared with the ratings to determine if there are any correlations between the acoustic characteristic and the trustworthiness ratings. After that, a neural network model will be built based on the collected data. The neural network model will be able to predict the trustworthiness of a person’s voice.

Keywords—prosody, trust, voice, vocal cues, zero acquaintance.