Sistema de identificación de locutor texto dependiente en Raspberry PI 3 B con aplicación en control de acceso.

Abstract

In this project, a recognition system of dependent text speaker using cepstral coefficients of the voice is developed using a Raspberry Pi 3 B with a USB audio card and a microphone with a 3.5mm jack connection. In this embedded device is implemented the capture of the audio signal, the signal conditioning, the extraction of the characteristics of the voice, the speaker recognition technique, the management of the database for each user of the system, the access registry of the users and the implementation of the interface. All system development is done in free software using the Python programming language in version 3.5, using the pyaudio library to capture the signal, mathematical libraries such as numpy, scipy, wave for signal conditioning, for extraction of the speech cepstral parameters the speech_features library is used, the recognition technique implemented is DTW and the database of the system is implemented in MySQL. In the end, tests are performed to identify false acceptances and false rejections and thus, determine the percentage of system error, its reliability and efficiency.