Signal processing and machine learning for smart sensing applications

• Autor(es): Chien, Ying-Ren ; Zhou, Mu ; Peng, Ao ; Zhu, Ni ; Torres-Sospedra, Joaquín
• Data: 2023
• Identificador Persistente: https://hdl.handle.net/1822/85317
• Origem: RepositóriUM - Universidade do Minho
• Tipo de Documento: Outro
• Idioma: Inglês
• Contribuidor(es): Universidade do Minho

Descrição

[Excerpt] The Special Issue “Signal Processing and Machine Learning for Smart Sensing Applications” focused on the publication of advanced signal processing methods by means of state-of-the-art machine learning technologies for smart sensing applications. It targeted research areas that included radio navigation, indoor/outdoor positioning, mm-wave sensing, speech denoising, and noise cancellation, among many others. A secondary objective was to promote interdisciplinary collaborations between researchers in the fields of signal processing and machine learning technologies for smart sensing applications. A total of 17 works were published within this Special Issue, where we can find works that are dealing with the more cutting-edge solutions for audio filtering for speech enhancement, identification and mitigation of some types of jamming, electroencephalogram processing for sleep-arousal detection, localization using magnetic field information, processing direction-of-arrival, detection of defects, fall detection, tracing healthcare data in real-time, as well as learn how signals propagate under non-line-of-sight conditions. The main contributions are briefly described in the remainder of this editorial. Zhou et al. [1] proposed a new algorithm using bone-conduction (BC) signals to assist dual-microphone generalized sidelobe canceller (GSC) adaptive beamforming for speech enhancement. First, the BC signals were used to conduct highly reliable voice activity detection (VAD), assisted adaptive noise canceller (ANC), and adaptive block matrix (ABM) weight coefficient updates in GSC. Second, an adaptive compensation filter (CF) was designed to compensate the amplitude and phase difference between air-conduction (AC) and BC signals. Third, wind noise was detected and replaced with the output of CF to recover low-frequency speech components from the wind noise. Finally, a real-time neural network-based postfilter was designed and trained to effectively remove the residual noise. Experimental results showed that the proposed algorithm effectively improves signal-to-noise ratio (SNR) and speech quality in different scenarios, and the assistance of BC signals can effectively improve the noise reduction performance of beamforming. [...]