A large-scale Wi-Fi channel state information dataset for contactless human speech recognition

Wi-Fi sensing has emerged as a promising contactless technology for understanding human activities and interactions by analyzing variations in wireless signals. In this vein, Wi-Fi-based spoken-word recognition from speech-associated movements offers a non-intrusive alternative for capturing speech-related information without microphones, supporting applications in privacy-preserving human-computer interaction, assistive technologies, and smart indoor environments. It is important to note that the Wi-Fi signals do not directly capture acoustic speech waves. Instead, the Channel State Information (CSI) and Received Signal Strength Indicator (RSSI) measurements reflect wireless-channel perturbations caused by speech-associated physical movements, including articulatory, facial, jaw, lip, and small head movements that occur during word pronunciation. This paper presents a dataset developed for Wi-Fi-based human speech recognition. The dataset consists of recordings of 30 distinct spoken English utterances, including single words and short phrases, collected from 25 subjects in an indoor environment. Each subject performed 30 trials for each utterance, resulting in a total of 22,500 trials across all participants, calculated as 25 subjects × 30 utterances × 30 trials. The data collection process used the publicly available CSI tool to capture Wi-Fi signals transmitted from a commercial off-the-shelf access point, the Sagemcom 2704, to a desktop computer equipped with an Intel 5300 network interface card. The recorded wireless measurements include both RSSI values and CSI values, providing complementary signal representations for speech-related analysis. The presented dataset provides a valuable resource for advancing research in Wi-Fi-based human speech recognition. It can support the development, evaluation, and comparison of machine learning and signal processing approaches for recognizing spoken English utterances from Wi-Fi signal variations. By providing a large and structured collection of trials from multiple subjects, the dataset enables further investigation into contactless speech sensing and contributes to the growing field of Wi-Fi-based human sensing.