Q-WiSE

Q-WiSE (Quantized Wiener-filter for Speech Enhancement) is Sensifai's flagship edge AI project developed under the dAIEdge Open Call #3. The system implements a quantized, AI-powered multi-channel Wiener filter optimised for deployment on microcontroller-class hardware operating at under 50 mW — making it suitable for hearing aids, smart earbuds, IoT voice interfaces, and industrial communication devices where cloud connectivity is absent or undesirable. By fusing classical signal-processing theory with modern neural network quantisation techniques (INT8 and INT4 weight precision), Q-WiSE achieves state-of-the-art noise suppression and speech intelligibility scores within severe power and memory constraints. The multi-channel beamforming module further enables directional noise rejection, isolating target speakers even in reverberant industrial environments. Q-WiSE represents the convergence of decades of Wiener-filter theory with the latest advances in quantized neural inference for deeply embedded systems.

Q-WiSE IS Sensifai's core EU-funded edge speech enhancement project. This project directly embodies the Q-WiSE programme's objectives — delivering ultra-low-power quantized neural inference for speech clarity at the edge, in line with the dAIEdge initiative's vision for intelligent edge devices.

This project supports several practical scenarios across research, commercialization, and deployment contexts:

• Environmental sound intelligence and classification
• Privacy-aware audio analytics pipelines
• Edge Speech Enhancement workflow automation
• dAIEdge Open Call #3 innovation validation
• Cross-functional product and research collaboration