In automated vehicles (AVs), the driver’s dual role as system supervisor and participant in Non-Driving Related Activities (NDRAs) poses a critical human-computer interaction challenge by competing for limited cognitive resources. While large language models have substantially advanced the semantic understanding of voice interaction, the mechanism aspect of how users initiate, maintain, and terminate commands remain underexplored in terms of usability and safety. To address this, our research systematically investigates Non-Verbal Auditory Inputs (NVAIs) and their integration with speech across a three-part experimental study. It progressed from foundational activation feasibility (Experiment 1, N=20) and single-task performance evaluation (Experiment 2, N=35) to quantifying cognitive interference within a dual-task driving simulator (Experiment 3, N=38). Our findings reveal critical trade-offs across the interaction stages. For activation (Experiment 1), users preferred the reliability of traditional wake-up words, whereas NVAIs received higher hedonic ratings but raised accessibility concerns. For continuous control (Experiment 2 and Experiment 3), a hybrid mechanism using a finger snap to terminate a spoken command significantly enhanced control precision without increasing interference to the primary supervisory task. Conversely, a mechanism mapping control to a sustained vowel extension proved unsuitable due to its high cognitive and physiological costs. In conclusion, these results demonstrate that the choice of input mechanism is critical under cognitive load. NVAIs such as snapping show promise as start-up and stop channels in supervisory AV contexts and motivate adaptive, multimodally redundant in-vehicle systems that better balance efficiency, user comfort, and safety.

This paper focuses on an interesting task, i.e., unsupervised cross-corpus Speech Emotion Recognition (SER), in which the labelled training (source) corpus and the unlabelled testing (target) corpus have different feature distributions, resulting in the discrepancy between the source and target domains. To address this issue, this paper proposes an unsupervised domain adaptation method integrating Transformers and Mutual Information (MI) for cross-corpus SER. Initially, our method employs encoder layers of Transformers to capture long-term temporal dynamics in an utterance from the extracted segment-level log-Mel spectrogram features, thereby producing the corresponding utterance-level features for each utterance in two domains. Then, we propose an unsupervised feature decomposition method with a hybrid Max-Min MI strategy to separately learn domain-invariant features and domain-specific features from the extracted mixed utterance-level features, in which the discrepancy between two domains is eliminated as much as possible and meanwhile their individual characteristic is preserved. Finally, an interactive Multi-Head attention fusion strategy is designed to learn the complementarity between domain-invariant features and domain-specific features so that they can be interactively fused for SER. Extensive experiments on the IEMOCAP and MSP-Improv datasets demonstrate the effectiveness of our proposed method on unsupervised cross-corpus SER tasks, outperforming state-of-the-art unsupervised cross-corpus SER methods.