GMM-driven recognition framework for canine whisperer identification

This paper is a dog language recognition system, aiming to study canine language recognition.The audio signals emitted by dogs are preprocessed and subjected to Mel Frequency Cepstral Coefficient (MFCC) feature extraction. Vector quantization is used to perform k-means clustering on the MFCC features of the dogs' daily audio. A Gaussian Mixture Model (GMM) is constructed, and the Expectation-Maximization (EM) algorithm is employed to train the GMM for each clustering center. By calculating the log-likelihood ratio scores between the test audio and the acoustic models of normal audio, the dog language recognition task is accomplished through score comparison.