Can AI Powered Speech-to-Text and Text-to-Speech techniques limit the interviewer bias in sensory and consumer research?

Abstract

Interviews using a human interviewer can be used to investigate product evaluations, however they can cause interviewer biases. Speech-to-Text and Text-to-Speech (StT&TtS) techniques can potentially reduce these biases. The study compared the difference in interviewer biases when a human interviewer is used or when AI Powered StT&TtS techniques are used. 53 panellists participated in two conditions: an interview with a human interviewer and a survey using StT&TtS techniques. In each condition two sweet snacks were evaluated: a regular variant (no health claim) and a new variant (same regular product but with health claim) and panellists were asked for their preference. Food consumption behaviour was also assessed. The new variant was rated higher in the human interview condition than in the StT&TtS condition (p = 0.001), the regular variant was not rated differently (p = 0.748). Within the human interviewer condition, the new variant was rated higher (p = 0.021) and more preferred (p < 0.001) than the regular variant. These effects were not seen within the StT&TtS condition (p = 0.782 and p = 0.492). Food consumption behaviour ratings did not differ between conditions (p = 0.685). In conclusion, using a human interviewer in sensory and consumer research can cause an interviewer bias, this bias can be limited by using Speech-to-Text and Text-to-Speech techniques.