Isolated Words Recognition Using a Low Cost Microcontroller

• Clayder Gonzalez-Cadenillas Universidad Nacional de Trujillo
• Nils Murrugarra-Llerena Universidad Nacional de Trujillo

Resumo

Currently, the field of automatic speech recognition is being widely used in commercial electronic devices such as TVs, phones, game consoles and computers. Thus, this article presents an evaluation of different isolated words recognition techniques on an embedded system using in the microcontroller dsPIC30F4013. So, the feature extraction phase is based on an adaptation of the Mel-frequency Cepstral Coefficients (MFCC) and the automatic recognition phase is based on the following techniques: Dynamic Time Warping (DTW), Artificial Neural Networks (ANN) and Principal Component Analysis (PCA). Related to the experiments setup, voice commands were evaluated in 3 different scenarios and the best accuracy rate was reached by a combination of PCA and ANN. It is also important to note that this implementation was carried out with the capacity constraints of the mentioned circuit.

Palavras-chave: Microcontrollers, Mel frequency cepstral coefficient, Artificial neural networks, Random access memory, Principal component analysis, Feature extraction, Signal processing algorithms, MFCC, DTW, ANN, PCA, dsPIC30F4013, isolated words recognition