Deep Speaker Embeddings for Speaker Verification of Children

Currently, deep speaker embedding models are the most advanced feature extraction methods for speaker verification. However, their effectiveness in identifying children’s voices has not been thoroughly researched. While various methods have been proposed in recent years, most of them concentrate on adult speakers, with fewer researchers focusing on children. This study examines three deep learning-based speaker embedding methods and their ability to differentiate between child speakers in speaker verification. The study evaluated the X-vector, ECAPA-TDNN, and RESNET-TDNN methods for forensic voice comparison using pre-trained models and fine-tuning them on children’s speech samples. The likelihood-ratio framework was used for evaluations using the likelihood-ratio score calculation method based on children’s voices. The Samromur Children dataset was used to evaluate the workflow. It comprises 131 h of speech from 3175 speakers aged between 4 and 17 of both sexes. The results indicate that RESNET-TDNN has the lowest EER and \( Cllr _{min}\) values (10.8% and 0.368, respectively) without fine-tuning the embedding models. With fine-tuning, ECAPA-TDNN performs the best (EER and \( Cllr _{min}\)  are 2.9% and 0.111, respectively). No difference was found between the sexes of the speakers. When the results were analysed based on the age range of the speakers (4–10, 11–15, and 16–17), varying levels of performance were observed. The younger speakers were less accurately identified using the original pre-trained models. However, after fine-tuning, this tendency changed slightly. The results indicate that the models could be used in real-life investigation cases and fine-tuning helps mitigating the performance degradation in young speakers.