2023/06/02 by Jinhan Wang, Wang, Jinhan, Vijay Ravi +3 · 1 citation
Computer Science · Medicine · Psychology · #Audio and Speech Processing (eess.AS) #FOS: Electrical engineering #Mental Health via Writing #Speech Recognition and Synthesis #Voice and Speech Disorders #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2306.01861
openalex publication_date 2023/06/02 · openalex created_date 2023/06/09 · openalex updated_date 2026/07/28
While speech-based depression detection methods that use speaker-identity features, such as speaker embeddings, are popular, they often compromise patient privacy. To address this issue, we propose a speaker disentanglement method that utilizes a non-uniform mechanism of adversarial SID loss maximization. This is achieved by varying the adversarial weight between different layers of a model during training. We find that a greater adversarial weight for the initial layers leads to performance improvement. Our approach using the ECAPA-TDNN model achieves an F1-score of 0.7349 (a 3.7% improvement over audio-only SOTA) on the DAIC-WoZ dataset, while simultaneously reducing the speaker-identification accuracy by 50%. Our findings suggest that identifying depression through speech signals can be accomplished without placing undue reliance on a speaker's identity, paving the way for privacy-preserving approaches of depression detection.