Sep 2026· Italian National Conference on Sensors· Vol 26, pp. 5794· 0 citations· 41 references
Medicine
TL;DR
An intelligent speech sensor is often evaluated at recognition output, although acquisition and preprocessing also determine what reaches control, so coverage, authorization, and acknowledged executor state need joint evaluation in an event-synchronized dual-frontend system.
Abstract
An intelligent speech sensor is often evaluated at recognition output, although acquisition and preprocessing also determine what reaches control. We examine this propagation in an event-synchronized dual-frontend system. A conditional cascade separates artifact availability, evidence formation, semantic conversion, and executor persistence. It decomposes how scheduled intent survives to the action boundary from raw audio through controlled motion. Paired recordings compare an INMP441–ESP32 interface with a USB interface. In an exploratory study, five speakers each delivered 30 Chinese utterances in three acoustic conditions, giving 450 scheduled events, 1800 ASR paths, and 3600 decision paths. Twelve missing artifacts remained in the intention-to-test denominators. Under the frozen peak normalization rule, isolated full-scale INMP samples limited window gain, and normalized USB–INMP speech-band differences reached +20.96 to +29.52 dB. Mean cloud text availability was 48.0 percentage points higher on USB. Exact valid execution rose by 41.7 points for DeepSeek and by 26.7 points for MiMo. Unauthorized non-stop motion on reject events rose by 19.6 points for DeepSeek and by 13.0 points for MiMo. Capability and risk increased together in all ten cloud speaker–consumer contrasts. Among 564 dangerous reject paths, 340 contained no literal motion token. In the worst case per speaker, hold-last replay produced 72.2–100% dangerous reject states. This is an exploratory sensor system case study of the tested frontend and preprocessing configuration, not a capsule ranking. Preprocessing changed useful evidence reach and action opportunity, so coverage, authorization, and acknowledged executor state need joint evaluation. The empirical results are specific to the tested pipeline and the five observed speakers.
This work frames this as Domain-Incremental Learning over acoustic environments and presents the first continual learning benchmark for multi-channel replay speech detection, evaluating a state-of-the-art beamformer-based detector over all 24 environment orderings of the ReMASC corpus with five seeds.
DF-VAD is introduced, an ASR-free streaming endpointer that maps bounded causal audio windows directly to Continue/Stop decisions and outperforms strong streaming and non-streaming semantic turn classifiers, and achieves the highest EOT recall among qualifying systems on TurnBench dev set $0.853.
Offline-trained speech enhancement models can degrade under deployment mismatch. Test-time adaptation (TTA) addresses this problem using unlabeled test observations, but existing approaches mainly rely on online parameter optimization and do not explicitly exploit the residual discarded by the enhancement backbone. We...
Y. Shan, Tian-Rui Wang, Zi-Xiang Wang et al.· 0 citations
Safety-critical control environments, such as nuclear power plant main control rooms, involve multi-operator collaboration, continuous information exchange, and stringent reliability requirements, making operator-state monitoring important for task performance and system safety. Speech is naturally produced, non-intrus...
AnomaSense is presented, a sensor activation approach for wrist wearables that keeps the microphone off by default and turns it on for at most one second when an unsupervised anomaly detector flags an IMU segment that is likely to produce sound.
Silent speech interfaces (SSIs) employing three-axis accelerometers can provide an effective communication modality for individuals who experience difficulty producing audible speech. However, the accuracy of such systems tends to degrade markedly when users move their heads (e.g., nodding or turning) during silent spe...
J. Sohn, Sung-Min Jung, Hye-seon Jeon et al.· IEEE Transactions on Audio,...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.