Audio Feature and Correlation Function-Based Speech Recognition in FM Radio Broadcasting

Abstract

The analysis and classification of audio signals are becoming increasingly important, especially in the age of communication and dissemination of information through radio broadcasting systems. It is therefore essential that systems and platforms are available to monitor the spread of fake or fraudulent news. A speech feature-based correlation (SFC) algorithm and a speech recognition framework are developed in this study, combining specific speech features and performance correlation to monitor real-time radio broadcasting and recognize specific speech based on human samples. The speech features include the Mel frequency cepstral coefficient, gammatone cepstral coefficient, spectral entropy, and pitch. The results illustrate the advantages and disadvantages of each feature applied to the various speech sound groups. Furthermore, each feature combined with the design of SFC further enhances system performance and increases accuracy.