{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/a-privacy-preserving-unsupervised-speaker","title":"A Privacy-Preserving Unsupervised Speaker Disentanglement Method for Depression Detection from Speech","arxiv_id":null,"date":"2024-02-28","proceeding":"AAAI ML4CMH Workshop 2024 2","authors":["Ravi","Vijay; Wang","Jinhan; Flint","Jonathan; Alwan","Abeer"],"abstract":"The proposed method focuses on speaker disentanglement in the context of depression detection from speech signals.\r\nPrevious approaches require patient/speaker labels, encounter instability due to loss maximization, and introduce unnecessary\r\nparameters for adversarial domain prediction. In contrast, the proposed unsupervised approach reduces cosine similarity\r\nbetween latent spaces of depression and pre-trained speaker classification models. This method outperforms baseline models,\r\nmatches or exceeds adversarial methods in performance, and does so without relying on speaker labels or introducing\r\nadditional model parameters, leading to a reduction in model complexity. The higher the speaker de-identification score\r\n(𝐷𝑒𝐼𝐷), the better the depression detection system is in masking a patient’s identity thereby enhancing the privacy attributes\r\nof depression detection systems. On the DAIC-WOZ dataset with ComparE16 features and an LSTM-only model, our method\r\nachieves an F1-Score of 0.776 and a 𝐷𝑒𝐼𝐷 score of 92.87%, outperforming its adversarial counterpart which has an F1-\r\nScore of 0.762 and 68.37% 𝐷𝑒𝐼𝐷, respectively. Furthermore, we demonstrate that speaker-disentanglement methods are\r\ncomplementary to text-based approaches, and a score-level fusion with a Word2vec-based depression detection model further\r\nenhances the overall performance to an F1-Score of 0.830.","url_abs":"https://ceur-ws.org/Vol-3649/Paper3.pdf","url_pdf":"https://ceur-ws.org/Vol-3649/Paper3.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"a-privacy-preserving-unsupervised-speaker","repo_url":"https://github.com/vijaysumaravi/USSD-depression","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"none","reach":null}],"tasks":[{"task_slug":"de-identification","task_name":"De-identification"},{"task_slug":"depression-detection","task_name":"Depression Detection"},{"task_slug":"disentanglement","task_name":"Disentanglement"},{"task_slug":"privacy-preserving","task_name":"Privacy Preserving"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":null,"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}