MUMBAI, India, Oct. 5 -- Intellectual Property India has published a patent application (202541032171 A) filed by Sandlogic Technologies Pvt. Ltd. on March 31, 2025, for An Efficient Speech-To-Text System For Low-Latency, Real-Time Transcription On Resource-Constrained Devices.
Inventors include Dr. Kruthika K R; and Syed Abdul Gaffar Shakhadri.
The application for the patent was published on October 02, 2026, under issue no. 40/2026.
Abstract: The present invention relates to an Automatic Speech Recognition (ASR) system configured for efficient speech-to-text processing using state-space modeling. The system comprises an audio preprocessing module for generating acoustic features, a feature extraction module for producing spectral representations, an encoder network comprising Selective State-Space Model (SSSM) processing blocks, and a decoder network comprising state-space residual blocks configured to generate textual output. A cross-connection module facilitates interaction between encoder and decoder using state-space transformations without employing attention mechanisms. In certain embodiments, the system includes a convolutional frontend for hierarchical feature extraction and temporal down sampling, a bidirectional encoder architecture, and a gating mechanism for dynamic feature fusion. The system achieves computational complexity proportional to sequence length, resulting in reduced latency and memory usage. The invention enables real-time speech recognition and efficient deployment on resource-constrained hardware platforms including edge devices and embedded systems. Fig. 1
Disclaimer: Curated by HT Syndication.
