Delving into VoxCeleb: environment invariant speaker recognition
This work addresses the issue of domain shift in speaker recognition for applications like biometrics, though it is incremental as it builds on existing adversarial training methods.
The paper tackled the problem of speaker recognition models learning environment-specific information by introducing an environment adversarial training framework that leverages video data from VoxCeleb to learn speaker-discriminative and environment-invariant embeddings, resulting in significant performance improvements on speaker identification and verification tasks.
Research in speaker recognition has recently seen significant progress due to the application of neural network models and the availability of new large-scale datasets. There has been a plethora of work in search for more powerful architectures or loss functions suitable for the task, but these works do not consider what information is learnt by the models, apart from being able to predict the given labels. In this work, we introduce an environment adversarial training framework in which the network can effectively learn speaker-discriminative and environment-invariant embeddings without explicit domain shift during training. We achieve this by utilising the previously unused `video' information in the VoxCeleb dataset. The environment adversarial training allows the network to generalise better to unseen conditions. The method is evaluated on both speaker identification and verification tasks using the VoxCeleb dataset, on which we demonstrate significant performance improvements over baselines.