Tag
Discusses how voice agents lose paralinguistic signals like tone, hesitation, and speaker identity when transcribing to text, and questions whether and how these features are captured and used downstream.
Analyzes filled pauses across 4 Slavic parliaments using transformer-based detection, replicating known effects (age, speech rate) and finding novel associations with sentiment, political orientation, and power status.
This paper introduces DEPOOL, a controlled benchmark evaluating six temporal aggregation architectures across six frozen speech backbones for depression detection in dyadic interactions, finding that many configurations collapse into single-class predictions and that robustness should be a key criterion.
Speech Playground is an interactive tool for speech analysis and comparison, combining a Python backend with a web frontend to explore multiple feature types and support utterance comparison for research and computer-aided pronunciation training.
This paper proposes a Multi-View Gated Graph Attention Network for Alzheimer's Disease detection from spontaneous speech, using semantic, dependency, and co-occurrence graphs with an adaptive gated fusion mechanism. The model achieves 90.00% accuracy on the ADReSSo dataset, and the source code is publicly available.
This paper presents a multi-stage explainable framework that combines SHAP-based token attribution, theory-informed linguistic features, and LLaMA-3.1-70B-Instruct LLM reasoning to interpret transformer-based speech models for cognitive impairment detection, achieving strong clinical alignment and high usability scores.
This paper investigates whether everyday speech from older adults can be used as a personalized cognitive monitoring tool, finding that AI models can detect subtle language patterns indicative of cognitive decline, unlike standard GPT responses.
Proposes a multimodal framework for fair Mild Cognitive Impairment detection from speech, using unlearning via gradient reversal to reduce demographic bias and improve performance across subgroups.
This paper introduces MA-DLE, a memory-based feature augmentation method for speech-based automatic depression level estimation, achieving state-of-the-art performance on the DAIC-WOZ and E-DAIC datasets.
This paper proposes a cross-linguistic transfer learning approach for detecting Alzheimer's Disease from speech across multiple languages, achieving F1 scores of 82% and supporting real-time screening applications.
This academic paper investigates using LLMs for zero-shot prediction of psychological well-being scores from spontaneous speech, evaluating 12 models and achieving high correlation with clinical metrics.