English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools annual

Global: Zendy Free Plan

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Preserving a patient's identity is a challenge for automatic, speech-baseddiagnosis of mental health disorders. In this paper, we address this issue byproposing adversarial disentanglement of depression characteristics and speakeridentity. The model used for depression classification is trained in aspeaker-identity-invariant manner by minimizing depression prediction loss andmaximizing speaker prediction loss during training. The effectiveness of theproposed method is demonstrated on two datasets - DAIC-WOZ (English) andCONVERGE (Mandarin), with three feature sets (Mel-spectrograms, raw-audiosignals, and the last-hidden-state of Wav2vec2.0), using a modified DepAudioNetmodel. With adversarial training, depression classification improves for everyfeature when compared to the baseline. Wav2vec2.0 features with adversariallearning resulted in the best performance (F1-score of 69.2% for DAIC-WOZ and91.5% for CONVERGE). Analysis of the class-separability measure (J-ratio) ofthe hidden states of the DepAudioNet model shows that when adversarial learningis applied, the backend model loses some speaker-discriminability while itimproves depression-discriminability. These results indicate that there aresome components of speaker identity that may not be useful for depressiondetection and minimizing their effects provides a more accurate diagnosis ofthe underlying disorder and can safeguard a speaker's identity.

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