A Step Towards Preserving Speakers' Identity While Detecting Depression Via Speaker Disentanglement

TL;DR

This paper introduces an adversarial disentanglement approach to improve depression detection from speech while preserving speaker identity, demonstrating effectiveness across multiple datasets and features.

Contribution

The study proposes a novel adversarial training method that disentangles depression features from speaker identity, enhancing classification accuracy and safeguarding speaker privacy.

Findings

Adversarial training improves depression classification across all features.

Wav2vec2.0 features with adversarial learning achieve the highest F1-scores.

Disentanglement reduces speaker discriminability while enhancing depression detection.

Abstract

Preserving a patient's identity is a challenge for automatic, speech-based diagnosis of mental health disorders. In this paper, we address this issue by proposing adversarial disentanglement of depression characteristics and speaker identity. The model used for depression classification is trained in a speaker-identity-invariant manner by minimizing depression prediction loss and maximizing speaker prediction loss during training. The effectiveness of the proposed method is demonstrated on two datasets - DAIC-WOZ (English) and CONVERGE (Mandarin), with three feature sets (Mel-spectrograms, raw-audio signals, and the last-hidden-state of Wav2vec2.0), using a modified DepAudioNet model. With adversarial training, depression classification improves for every feature when compared to the baseline. Wav2vec2.0 features with adversarial learning resulted in the best performance (F1-score of 69.2% for DAIC-WOZ and 91.5% for CONVERGE). Analysis of the class-separability measure (J-ratio) of the hidden states of the DepAudioNet model shows that when adversarial learning is applied, the backend model loses some speaker-discriminability while it improves depression-discriminability. These results indicate that there are some components of speaker identity that may not be useful for depression detection and minimizing their effects provides a more accurate diagnosis of the underlying disorder and can safeguard a speaker's identity.