Desarrollo de un algoritmo de reconocimiento de palabras mediante un microcontrolador para GIECA

Abstract

In this project a prototype is studied that is capable of recognizing specific words that are heard through a microphone, a microcontroller and a sound response according to the dictated word. To achieve the aforementioned objective, an electronic prototype for voice recognition has been implemented using a KPU K210 microcontroller as the central processor of the Maix Go electronic circuit where 2 programming methods are stored, the first one will only record the responses that will be reproduced in the loudspeaker. From the prototype stored these audios on a 64 gigabyte SD card, the second method will record the voice commands that will be compared and will extract a cepstral vector with its own characteristics, to compare it with the voice commands that the speaker recognizes that Like the commands previously stored in memory, it will extract characteristics of each one and determine a cepstral vector to compare it. To sense spoken words, a highly sensitive I2S APU (audio processor) audio module is processed, which will be connected to one input of the Maix Go prototype. Once the audio signal is processed it emits an audible response through the IMAFDC RISC-V player module by sending a signal from the microcontroller via serial communication. In the same way that it sends a signal through the serial port to the audio player, it will send several a signals to a PC to present the text response of the recognized word.