Exploring racial and gender disparities in voice biometrics

Abstract

Systemic inequity in biometrics systems based on racial and gender disparities has received a lot of attention recently. These disparities have been explored in existing biometrics systems such as facial biometrics (identifying individuals based on facial attributes). However, such ethical issues remain largely unexplored in voice biometric systems that are very popular and extensively used globally. Using a corpus of non-speech voice records featuring a diverse group of 300 speakers by race (75 each from White, Black, Asian, and Latinx subgroups) and gender (150 each from female and male subgroups), we explore and reveal that racial subgroup has a similar voice characteristic and gender subgroup has a significant different voice characteristic. Moreover, non-negligible racial and gender disparities exist in speaker identification accuracy by analyzing the performance of one commercial product and five research products. The average accuracy for Latinxs can be 12% lower than Whites (p < 0.05, 95% CI 1.58%, 14.15%) and can be significantly higher for female speakers than males (3.67% higher, p < 0.05, 95% CI 1.23%, 11.57%). We further discover that racial disparities primarily result from the neural network-based feature extraction within the voice biometric product and gender disparities primarily due to both voice inherent characteristic difference and neural network-based feature extraction. Finally, we point out strategies (e.g., feature extraction optimization) to incorporate fairness and inclusive consideration in biometrics technology.

Programmable Polymeric‐Interface for Voiceprint Biometrics

“…Females generally have higher pitch while speaking with a shorter duration, whereas men generally possess larger and thicker vocal cords and may produce louder and longer speech more easily. [ 50 ]…”

“…Females generally have higher pitch while speaking with a shorter duration, whereas men generally possess larger and thicker vocal cords and may produce louder and longer speech more easily. [50] Principal component analysis (PCA) has been also carried out based on the voiceprint for classifying the male/female population (Figure S16, Supporting Information) and a pair plot demonstrates the relation among all the features (Figure S17, Supporting Information). [51] With the aim to recognize the classification capabilities of the PAS for classifying the male and female population, a confusion matrix has been deployed further, that indicates the population has been detected with an accuracy of >96% (Figure 4g).…”

To train or not to train adversarially: A study of bias mitigation strategies for speaker recognition

“…A recent work explored racial and gender disparities in several speaker classification systems [75]. However, their analysis was restricted to closed-set classification task, which is different from the more challenging ASV setup we consider in this work.…”

Building an Arabic Dialectal Diagnostic Dataset for Healthcare

“…Finally, most of the participants ( 80%) are from the young generation, thereby minimizing the contribution of children, middle age and elder samples of the Emirati and Egyptian populations. This systemic inequity based on age and gender disparities has received a lot of attention recently in voice biometrics [76] and reveals that age and gender subgroups have a significant different voice characteristic. In fact, non-negligible gender disparities exist in speaker identification accuracy and show that the average accuracy can be significantly higher for female speakers than males due to mainly voice inherent characteristic difference [76].…”

“…This systemic inequity based on age and gender disparities has received a lot of attention recently in voice biometrics [76] and reveals that age and gender subgroups have a significant different voice characteristic. In fact, non-negligible gender disparities exist in speaker identification accuracy and show that the average accuracy can be significantly higher for female speakers than males due to mainly voice inherent characteristic difference [76]. The results in [77] also indicate a significant age effect on the voice acoustic parameters (fricative spectral center of gravity, spectral skewness, and speaking STSD) revealing that certain speech and voice features change with age.…”