MUMBAI, India, Aug. 10 -- Intellectual Property India has published a patent application (202411105237 A) filed by Amity University on December 31, 2024, for Res42net-Transformative Psychoacoustic Scale-Based Text-Independent Speaker Recognition System.
Inventors include Pooja Gambhir; and Dr. Amita Dev.
The application for the patent was published on July 31, 2026, under issue no. 31/2026.
Abstract: The present invention discloses a Res42NEt-Transformative psychoacoustic scale-based Text-Independent Speaker Recognition System for identifying speakers irrespective of spoken content. The system employs a deep residual neural network architecture configured with five residual blocks to process psychoacoustic feature representations extracted directly from raw audio signals. Mel-Filter Cepstral Coefficients, Sinc Band-Pass Filter features, and Bark-Frequency Psychoacoustic features are utilized to enhance speaker discrimination under short-duration utterances and low signal-to-noise ratio conditions. The architecture incorporates one-dimensional convolutional layers with skip connections, statistical pooling for fixed-dimensional embedding generation, and SoftMax-based classification trained using sparse categorical cross-entropy loss. The system demonstrates improved speaker identification accuracy and reduced equal error rate compared to baseline residual network architectures.
Disclaimer: Curated by HT Syndication.
