Improving Voice Identification Procedures Experimental Data, 2019-2024

The Improving Voice Identification Procedures (IVIP) project consisted of three stands: Strand 1 - Parameters, Strand 2 - Voice Distinctiveness and Strand 3 - Social Stereotypes. All data has been archived in this collection.

Strand 1: Parameters - Motivation and aims
This strand of the project examined aspects of voice identification procedure which were yet to be fully tested within the Home Office approach, such as the length of voice samples, number of foil voices, witness instructions, and parade type. A series of experiments was conducted to identify optimal parameters to increase both the chances that guilty suspects are selected, and that innocent suspects are not selected.

Strand 1: Parameters - Key findings
Experiments (1 and 2) testing 15s, 30s, and 60s voice samples showed 15s can be used without adversely affecting witness identification performance. Experiments (1 and 2) investigating parade size found no meaningful difference in performance for six- versus nine-person parades.

A study considered the impacts of reflecting on a voice after exposure to it (Experiment 3), and shorter (~5 minutes) versus longer (~1 day) retention intervals. No meaningful difference in identification performance was found between reflection/no-reflection conditions, nor shorter/longer retention intervals.

A study testing different warning strengths in witness instructions (Experiments 4 and 5) showed stronger warnings improve accuracy in target-absent parades, but at the expense of target-present accuracy. Analysis of data comparing serial and sequential parade formats both online (Experiments 4, 5 and 6) and in the lab (Experiment 6) did not reveal differences in performance across parade types or testing contexts.

Experiment 7 addressed the effect of target-foil similarity and the opportunity to relisten to parade voices. Preliminary analyses suggest no difference in performance according to target-foil similarity, but higher accuracy on target-present parades when listening to parade voices once.

Strand 2: Voice Distinctiveness - Motivation and aims
Voice parade experiments tend to show that certain target speakers are more readily identifiable than others. This strand of the project considered what makes some speakers more distinctive-sounding than others, and whether speakers judged to be more distinctive-sounding are also more memorable. Experiments were conducted to explore correlates of judgements of voice similarity and phonetic features in combination with accent. Further experiments investigated whether the distinctiveness of a voice depends on its position in a population distribution of voices according to long-term f0, a phonetic feature known to be perceptually important for speaker identity, and the relationship between this factor and voice memorability.

Strand 2: Voice Distinctiveness - Key findings
Varying patterns of correlation were found between listener-judged voice similarity and pitch, formants and articulation rate, within and between different accent groups, with pitch and formants playing important roles (Experiment 1). An experiment exploring the effect of sample duration (3s vs 10s) on listener-judged voice similarity found similar correlations across sample durations (Experiment 2). Further experiments (3, 4, and 5) have examined the role of pitch in listeners’ judgement of voice distinctiveness. Using pitch-manipulated stimuli, the study highlighted that listeners hear the same pairs of speakers as more different in the mid-range than when heard low or high (Experiment 4). Further experimentation showed this was because listeners are more sensitive to pitch differences in the mid-range (Experiment 5). A further experiment (Experiment 6) found that the benefit of mid-pitch on perceived distinctiveness did not extend to memorability, but instead found that voices that are distinctive for reasons other than pitch were more memorable for listeners.

Strand 3: Social Stereotypes - Motivation and aims
Strand 3 examined the extent to which social perceptions, judgements, attitudes and stereotypes related to voice(s) can motivate witness decision-making during voice parades. Three experiments assessed the relationship between certain aspects of voice and stereotypes about traits and particular criminal and non-criminal behaviours. Developing an understanding of the link between voice stereotypes and voice identification has the potential to reduce the likelihood of an innocent person being selected from a line-up because that person happens to have the most ‘guilty’ sounding voice.

Strand 3: Social Stereotypes - Key findings
Experiment 1 showed that listeners consistently rated Standard Southern British English (SSBE) highest for all status traits (concerning education, wealth, intelligence) while Belfast, Cardiff and Glasgow rated highest on the solidarity traits (friendliness, trustworthiness, honesty). Belfast and Glasgow also rated well on ‘honourable’ behaviours and low on ‘morally bad’ behaviours. SSBE speakers were thought least likely to commit criminal behaviours and Liverpool speakers most likely.

Experiment 2 showed that low-pitched voices were rated lower for solidarity-based traits, whereas high-pitched voices were rated lower for status-based traits. Slower articulation rates resulted in lower ratings for status and solidarity traits and morally good behaviours, but higher ratings for criminal behaviours.

Experiment 3 examined whether stereotyping in voice judgements could contribute to (mis)identification of those speakers in voice parades. It found voices most frequently selected from parades were rated more negatively regarding social and behavioural traits (including criminality).