International Conference on Computer and Knowledge Engineering

Home / 13th International Conference on Computer and Knowledge Engineering

AgeNet-AT: An End-to-End Model for Robust Joint Speaker Age Estimation and Gender Recognition Based on Attention Mechanism and Titanet

Authors :

Mahsa Zamani Tarashandeh¹ Amirhossein Torkanloo² Mohammad Hossein Moattar³

1- Department of Electrical Engineering, Faculty of Engineering Ferdowsi University of Mashhad 2- Department of Computer Engineering Ferdowsi University of Mashhad Mashhad, Iran 3- ,Department of Computer Engineering Mashhad Branch, Islamic Azad University, Mashhad, IRAN

Keywords :

Age estimation،Gender classification،Multi-task learning،Attention mechanism،Titanet

Abstract :

Speaker age estimation has become popular in recent years due to its potential applications in various fields, including forensics and human-computer interaction. However, noise and utterance length robustness is a key factor in the performance of the approaches. In this work, a robust age estimation and gender recognition model named AgeNet-AT is proposed based on an attention mechanism and Titanet model. The proposed approach applies Titanet as the embedding extractor, and attention mechanism to create an end-to-end architecture for age estimation. Since Titanet is a model designed to distinguish different speaker identities, it is hypothesized that some of its extracted features may contain properties that can differentiate speakers’ age and gender. Therefore, Titanet is chosen as the embedding approach in this study. Additionally, an attention layer is used to focus on the most valuable features for age estimation. Furthermore, an auxiliary task of gender classification is added to the model in order to improve the estimation performance. The experiments are conducted on TIMIT dataset for different evaluation conditions, such as various utterance lengths and noise levels. The experimental results indicate the robustness of the AgeNet-AT model. The model has outperformed the state-of-the-art age estimation results on TIMIT dataset with Root Mean Square Error (RMSE) of 5.92 and 6.85 and Mean Absolute Error (MAE) of 4.30 and 4.73 for male and female speakers, respectively.