MFCC for Voiced Part Using VAD and GMM Based Gender Recognition

Abstract

For many applications, identifying the gender information of a speaker is important. In this paper, we implemented the system which identifies the speaker and also gender of the speaker by using MFCC and GMM in an uncontrolled environment. In this text independent system, we aim on the classification using GMM for the extracted features using MFCC and also the speech signal is processed with Voice Activity Detector (VAD). In the experiments using locally recorded database, the system without voice activity detector (VAD) does not provide accurate results. So, the main aim of this paper is to develop a text independent speaker identification and also gender identification using MFCC along with VAD and GMM which improves the performance further relatively when compared with the system without VAD. The performance of the proposed system tested for 70 speakers with 100 percent recognition rate is achieved based on the log likelihood scores.

A Parallel Classification Model for Marine Mammal Sounds Based on Multi-Dimensional Feature Extraction and Data Augmentation

“…Compared to these traditional recognition algorithms, machine learning based on a large number of data samples can fit a very complex and accurate prediction function, so it is increasingly used in the field of acoustic detection. Typical machine learning based audio recognition algorithms comprises decision trees [ 35 ], linear discriminant analysis (LDA) [ 36 ], support vector machines (SVMs) [ 37 ], the Gaussian mixture model (GMM) [ 38 ], self-organizing maps (SOMs) [ 39 ], long short-term memory (LSTM) [ 40 ], the hidden Markov model (HMM) [ 41 ], and convolutional neural networks (CNNs) [ 42 , 43 , 44 , 45 , 46 ]. As computers become more powerful, researchers are processing larger amounts of data.…”