Assessing the abilities of phonetically untrained listeners to determine pitch and speaker accent in unfamiliar voices

Abstract

It is sometimes the case that a victim of a crime will never see the perpetrator’s face, but will be exposed to his or her voice. This could occur in situations such as masked robberies, telephone fraud, or the receipt of bomb threats via phone or voicemail. In such cases, attempts can be made by the police or intelligence services to get the witness or victim to describe the voice of the offender. However, there is a high likelihood that a given earwitness will lack the linguistic expertise and technical vocabulary of the kind used by trained phoneticians when they describe voices. One question that arises from this problem is whether phonetically untrained listeners have the ability, using verbal means, to accurately capture different aspects of speakers’ voices. This paper presents an experiment in which a group of listeners were tasked with assessing how ‘high-pitched’ the voices of 12 speakers were, along with providing a description of each speaker’s accent. These assessments were then compared to measured Fundamental Frequency (F0) values and prior knowledge of the speakers’ accents in order to assess listener performance. The results suggest that while some listeners have the ability to make reliable judgements of relative vocal pitch, the overall correlations between measured (F0) and perceived pitch were weak. With regard to accent, the results suggest that the more unfamiliar a speaker’s accent is to the listener, perhaps owing to the geographical distance of the area where the accent is spoken from the listener’s place of origin, the harder it will be for the listener to accurately describe that accent. We argue that testing the abilities of earwitnesses to assess aspects of speakers’ voices before their descriptions are used further would be a useful safeguard against the use of potentially inaccurate or erroneous earwitness evidence in police investigations.