A lightweight speaker verification approach for autonomous vehicles

Speaker verification is the process of verifying an individual’s identity by comparing their recorded voice samples with their test speech signals. Speaker verification has various practical applications, such as verifying customer identities in call centers, enabling contactless facility access, and supporting some medical applications. With the advances in autonomous vehicles, speaker verification has become an essential feature that provides security, access control, personalization, command authentication, driver monitoring, and compliance. Recent technological advancements have led to the rise of voice-based authentication systems, which are considered a more convenient alternative to traditional security systems. However, improving the accuracy is still an ongoing research aim. In this research, four different models were proposed and compared with previous work on speaker verification. The models are combinations of using two networks (BiLSTM and Transformer) with two different loss functions (Triplet and Quadruplet loss functions). The models are trained and tested on the LibriSpeech dataset. The results show improvements in equal error rate of the four proposed models over the previous models that used the Librispeech dataset with 0.068 compared to 0.11. Received: 29 October 2024 Accepted: 03 December 2024 Published: 22 December 2024