The Jena Voice Learning and Memory Test (JVLMT): A standardized tool for assessing the ability to learn and recognize voices

Abstract

The ability to recognize someone's voice spans a broad spectrum with phonagnosia on the low end and super-recognition at the high end. Yet there is no standardized test to measure an individual's ability of learning and recognizing newly learned voices with samples of speech-like phonetic variability. We have developed the Jena Voice Learning and Memory Test (JVLMT), a 22-min test based on item response theory and applicable across languages. The JVLMT consists of three phases in which participants (1) become familiarized with eight speakers, (2) revise the learned voices, and (3) perform a 3AFC recognition task, using pseudo-sentences devoid of semantic content. Acoustic (dis)similarity analyses were used to create items with various levels of difficulty. Test scores are based on 22 items which had been selected and validated based on two online studies with 232 and 454 participants, respectively. Mean accuracy in the JVLMT is 0.51 (SD = .18) with an empirical (marginal) reliability of 0.66. Correlational analyses showed high and moderate convergent validity with the Bangor Voice Matching Test (BVMT) and Glasgow Voice Memory Test (GVMT), respectively, and high discriminant validity with a digit span test. Four participants with potential super recognition abilities and seven participants with potential phonagnosia were identified who performed at least 2 SDs above or below the mean, respectively. The JVLMT is a promising research and diagnostic screening tool to detect both impairments in voice recognition and super-recognition abilities.

Keywords: IRT; Individual differences; Phonagnosia; Voice recognition.

Fig. 1

Acoustic voice space containing all voice triplets of an exemplary male learning voice identity who is here represented as “target voice”. Colored triangles represent different levels of acoustic similarity between the target and two respective foils, as used on a given test trial. Triplets used for immediate recognition trials in the learning phase had medium acoustic similarity

Fig. 2

Overview of the development steps (1) and the JVLMT paradigm (2a–c) as exemplified for one learning voice identity (shown in teal). The JVLMT consists of three phases: The learning phase (a) consists of a familiarization trial followed by an immediate recognition trial comprising the learned target voice and two foils (shown in red and pink). In the repetition phase (b), participants listen to the learning voice identities who repeat a sentence from the familiarization trial and from the immediate recognition trial of the learning phase. In the testing phase (c), participants perform a 3AFC-task in which they have to recognize the previously learned voice identities among two new foils (shown in green and blue). Duration of utterances varied (i.e., 250 ms, 750 ms, full duration) between trials as symbolized by the number of speech output lines. Speech content varied between familiarization and test in the learning phase and also between the learning and testing phase as represented by varying colors of speech output lines. The figure was designed with Canva (Canva, ; Gehred, 2020)

Fig. 3

Proportion of correct responses in the testing phase. Depicted factors are duration (250 ms, 750 ms, and full length) and acoustic similarity (low, medium, and high) of test voices on a given trial. Chance level (0.33) is indicated by the dotted line. Error bars represent SEMs

Fig. 4

Item-characteristic curves (1) for the 22 items conforming to the Rasch model from the final version of the JVLMT. (2) contains the test information function (blue line) and the standard error of ability levels (pink line)

Fig. 5

Mean proportion of correct responses for each test item, depicted in chronological order of appearance in the JVLMT, averaged over N = 454 online data sets. The dashed line represents the chance level (0.33). Error bars represent SEMs

Fig. 6

Test score distribution of the JVLMT. Dashed lines represent 2 SDs below/above the mean

Fig. 7

Test score distribution of the JVLMT divided by age groups (1) and gender (2)