MUMBAI, India, Sept. 22 -- Intellectual Property India has published a patent application (202641110564 A) filed by National Institute Of Technology-Warangal on September 15, 2026, for Hybrid Lstm Model For Speech Command Recognition Using Edge Impulse.
Inventors include Mirthipati Saikiran Sanjay; and Prof. Ravi Kumar Jatoth.
The application for the patent was published on September 18, 2026, under issue no. 38/2026.
Abstract: TITLE: Hybrid LSTM Model for Speech Command Recognition Using Edge Impulse The present invention discloses a real-time speech command recognition system and method for low-power embedded edge devices, exemplified by the Arduino Nano 33 BLE Sense as shown in figure-1. The system captures audio signals through a microphone and extracts Mel-Frequency Cepstral Coefficients (MFCCs), including delta and delta-delta derivatives, to represent temporal speech dynamics. These features are processed through a one-dimensional convolutional neural network (1D CNN) for spatial pattern extraction, followed by a custom long short-term memory (LSTM) block that models sequential temporal dependencies. The hybrid CNN–LSTM model is trained and quantized to reduce memory usage and computational complexity, enabling efficient on-device inference. Dense and softmax layers classify the processed features into predefined speech commands, which are then displayed and used to trigger corresponding actions via an output interface. The system operates fully offline, achieving low-latency, accurate recognition while remaining within the strict RAM and Flash memory constraints of embedded edge devices.
Disclaimer: Curated by HT Syndication.
