Меню
Vol. 4 No. 29 (2025) / 2025-12-30 / Number of views: 204
Authors
Keywords
Link to DOI:
How to quote
A neural network-based method is proposed for extracting speaker articulation characteristics. It employs specialized trigram statistics to obtain a representative spectrogram that provides a stable representation of the speaker’s articulation through a linear superposition of short-term spectrograms. To address the issues of slow training and recognition speed in speaker recognition systems on resource-constrained devices, an Adaptive Clustering Self-Organizing Map (AC-SOM) algorithm is introduced, based on the traditional Self-Organizing Map (SOM) neural network. This algorithm automatically adjusts the number of neurons at the competitive layer based on the number of speakers to be recognized, until the number of clusters matches the number of speakers. A dataset comprising characteristic spectrogram patterns from 100 speakers was developed and used for the proposed AC-SOM model, achieving a maximum training time of only 304 seconds and a maximum recognition time of less than 28 milliseconds per sample. Compared to other approaches, the proposed method significantly improves training and recognition speed without sacrificing high recognition accuracy. The promising results indicate that, in contrast to existing speaker recognition techniques, the proposed method better meets the real-time data processing and execution requirements of edge intelligence systems.