Implementação em FPGA de um Sistema Keyword Spotting Usando Redes Neurais Convolucionais

Natan Votre, Walter Gontijo, Eduardo L. O. Batista

DOI: 10.14209/SBRT.2020.1570660547

Evento: XXXVIII Simpósio Brasileiro de Telecomunicações e Processamento de Sinais (SBrT2020)

Keywords: keyword spotting FPGA CNN MFCC

Abstract

This paper presents the field-programmable-gate-array (FPGA) implementation of a keyword spotting (KWS) system focused on the detection of specific words in Portuguese language. The implemented system is based on the use of mel-frequency cepstral coefficients (MFCCs) as features for a convolutional neural network (CNN). Initially, the proposed KWS system is implemented and evaluated using the Python programming language along with the TensorFlow module. Afterwards, the FPGA synthesis of the proposed system is carried out as well as its evaluation using an appropriate development kit. The obtained results show that the implemented system is very effective and capable of processing dozens of audio channels in real time.

Download