CONTRIBUTION TO THE DEVELOPMENT OF AN AUDITORY MODULE FOR ASSISTIVE ROBOTICS
Artificial Neural Networks; Descriptor Extraction; MFCC; Automatic Speaker Recognition; Automatic Emotion Recognition; Independent Component Analysis;
This work proposes an automatic system of recognition of speaker and emotions through the voice signal with the application of statistical signal processing techniques such as the principal components analysis and the independent components analysis and artificial neural classifiers. Initially, characteristic descriptors of audio and voice signals such as Mel frequency cepstral coefficients are extracted from the audio files available in the used dataset. After this step, principal component analysis and independent component analysis techniques are applied to reduce dimensionality and remove redundant information between the parameters that will be presented at the input of the shallow neural network. In order to compare the results, also, pre-trained convolutional neural networks and recurrent neural networks with memory cell LSTM and BLSTM are used and that pre-trained convolutional neural classifiers, in an image database, can be used for the automatic recognition of emotions and speakers task achieving good discrimination efficiency.