Speech-Based Depression Prediction Using Encoder-Weight-Only Transfer Learning and a Large Corpus

Abstract

Speech-based algorithms have gained interest for the management of behavioral health conditions such as depression. We explore a speech-based transfer learning approach that uses a lightweight encoder and that transfers only the encoder weights, enabling a simplified run-time model. Our study uses a large data set containing roughly two orders of magnitude more speakers and sessions than used in prior work. The large data set enables reliable estimation of improvement from transfer learning. Results for the prediction of PHQ-8 labels show up to 27% relative performance gains for binary classification; these gains are statistically significant with a p-value close to zero. Improvements were also found for regression. Additionally, the gain from transfer learning does not appear to require strong source task performance. Results suggest that this approach is flexible and offers promise for efficient implementation.

Feasibility of a Machine Learning-Based Smartphone Application in Detecting Depression and Anxiety in a Generally Senior Population

“…Usage statistics and survey results from this study taken together indicate that utilizing regular voice recordings of users answering questions from a smartphone app to analyze their levels of anxiety and depression is feasible. Ellipsis Health has previously published results of semantic ( Rutowski et al, 2019 , 2020 ) and acoustic ( Harati et al, 2021 ) analysis of speech to detect depression and anxiety using models trained, to the best of our knowledge, with the largest database reported in the literature ( Rutowski et al, 2019 ). We have also previously reported this algorithm performance is maintained (i.e., is portable) when applied to the current study population using long short-term memory (LSTM) models ( Rutowski et al, 2020 ).…”

“…In ongoing work, large corpora of transcribed speech are used for natural language processing (NLP) training to further develop semantic speech analysis ( Lan et al, 2019 ; Yang et al, 2019 ). The most relevant NLP advancements have been used for transfer learning and improvements in deep learning architecture like transformers ( Bengio, 2012 ; Vaswani et al, 2017 ), which maintain performance without using prohibitive amounts of labeled data ( Rutowski et al, 2020 ; Harati et al, 2021 ).…”

Evaluating the Feasibility and Acceptability of an Artificial-Intelligence-Enabled and Speech-Based Distress Screening Mobile App for Adolescents and Young Adults Diagnosed with Cancer: A Study Protocol

“…In addition, the platform offers an oncologist-facing dashboard, which facilitates patient referral for psycho-oncology services and allows timely coordination of care and patient-/person-centered approaches. Finally, Ellipsis Health has published a series of peer-reviewed technical papers validating the machine learning algorithms as well as the speech recognition performance that power the approach [ 39 , 40 , 41 , 42 , 43 ].…”

FrAUG: A Frame Rate Based Data Augmentation Method for Depression Detection from Speech Signals

“…Recently, speech-based automatic diagnosis of depression has gained significant momentum [6,7,8] and advancements in deep learning have pushed their performance to newer heights [9,10,11,12,13,14]. However, data scarcity still remains one of the major challenges in building reliable systems for MDD modeling purposes.…”