AUTHOR=Cavalcanti Julio Cesar , da Silva Ronaldo Rodrigues , Eriksson Anders , Barbosa Plinio A. 

TITLE=Exploring the performance of automatic speaker recognition using twin speech and deep learning-based artificial neural networks

JOURNAL=Frontiers in Artificial Intelligence

VOLUME=Volume 7 - 2024

YEAR=2024

URL=https://www.frontiersin.org/journals/artificial-intelligence/articles/10.3389/frai.2024.1287877

DOI=10.3389/frai.2024.1287877

ISSN=2624-8212

ABSTRACT=This study explored the impact of speaker similarity and sample length on the performance of an automatic speaker recognition (ASR) system based on the SpeechBrain toolkit. The research used speech materials collected from 20 male identical twin speakers in two speaking styles: spontaneous dialogues and interviews. The study included same-speaker tests that use data from the same subject in mismatched speaking styles. The ASR performance was evaluated by comparing identical twins, all speakers in the corpus including twin pairs, and all speakers excluding twin pairs. The speech samples evaluated varied from 5 s to 30 s. The system's performance was assessed using equal error rates (EER) and Log cost-likelihood ratios (Cllr). The results indicated that identical twins pose a considerable challenge to the ASR process, reducing the overall speaker recognition accuracy. Moreover, ASR analyses relying on longer speech samples tended to outperform comparisons involving shorter samples. Additionally, with an increase in sample size, standard deviation values for both intra and inter-speaker similarity scores decreased. This suggests a reduced variability in estimating speaker similarity/dissimilarity levels in longer speech segments compared to shorter ones. The findings also revealed that among identical twins, there were varying levels of likeness. This variability in the level of likeness between twin speakers could imply that some individuals pose a greater challenge for ASR systems. The outcomes are compatible with previous findings and are discussed in light of the relevant literature.