Search papers, labs, and topics across Lattice.
This study introduces DEPOOL, a benchmark for evaluating temporal aggregation strategies in speech-based depression detection, addressing the limitations of existing methods that fix encoder layers and overlook the impact of aggregation architectures. By systematically comparing six aggregation architectures across six frozen speech backbones on English and Mandarin depression corpora, the authors reveal that many configurations fail by defaulting to a single-class prediction, highlighting the critical influence of both the backbone and the aggregation method. The findings emphasize the need for robustness in benchmarking, as the most stable architectures in single-seed runs proved unreliable across multiple training seeds.
A third of tested configurations in depression detection collapsed to a single-class prediction, underscoring the hidden pitfalls of current aggregation methods.
Speech-based depression detection compresses features from short audio segments into one speaker-level decision, a step called temporal aggregation rarely studied on its own. Most benchmarks fix a single self-supervised encoder and a single hand-picked layer, so a reported gain may reflect the pipeline rather than the aggregation method itself. We introduce DEPOOL, a controlled benchmark that compares six aggregation architectures with six frozen speech backbones on an English and a Mandarin depression corpus, where each configuration learns which backbone layers matter rather than fixing one by hand. Across the resulting 72-configuration grid, a third of configurations collapse into predicting a single class for every speaker, a failure tied to the backbone as much as to the method, and the architecture that is most stable in a single-seed run becomes unreliable when training repeats across seeds. Robustness to backbone and seed, rather than average accuracy across a single pipeline, should be a first-class benchmarking criterion for temporal aggregation in clinical speech.
