MUMBAI, India, July 24 -- Intellectual Property India has published a patent application (202611060002 A) filed by Graphic Era Hill University, Dehradun on May 12, 2026, for A System And Method For On-Device Voice Activity Detection Using A Digital Mems Microphone With Sub-10 Millisecond Latency.
Inventors include Simar Singh Rayat; Sujal Thapa; Raj Basnet; Dr. Vikrant Sharma; and Dr. Arun Pratap Singh Rathod.
The application for the patent was published on July 17, 2026, under issue no. 29/2026.
Abstract: ABSTRACT A SYSTEM AND METHOD FOR ON-DEVICE VOICE ACTIVITY DETECTION USING A DIGITAL MEMS MICROPHONE WITH SUB-10 MILLISECOND LATENCY The invention relates to on-device voice activity detection in embedded systems and employs a digital MEMS microphone module with an integrated sigma-delta modulator interfaced via an I²S synchronous bus to a single-board computer. At boot, a self-calibration module measures ambient RMS audio energy, adjusts gain, corrects DC offset, aligns bit-clock and word-select phases, and stores calibration parameters. A DMA controller transfers serialized audio into a lock-free circular buffer sized for up to twenty milliseconds. Overlapping ten-millisecond frames undergo adaptive noise modeling, STFT, and spectral gain processing. A feature extraction subsystem computes log-mel spectrograms, temporal derivatives, zero- crossing rate, and energy metrics. An INT8-quantized neural inference core with runtime pruning produces posterior probabilities, which decision logic thresholds to declare voice presence. Dynamic voltage and frequency scaling optimizes power while ensuring end-to-end latency below ten milliseconds.
Disclaimer: Curated by HT Syndication.
