1R-08
FFTアクセラレーションを活用したRISC -VベースのMFCC音声認識に関する研究
Mel-Frequency Cepstral Coefficients (MFCC) is a technique used to obtain a signal's power spectrum for recognition applications. The Mel-coefficients are used in the different speech recognition methods, obtaining a high recognition rate. However, the most processing time is used for the Fast Fourier Transform (FFT). This work presents an FFT accelerator implemented in Field-Programmable Gate Array (FPGA), using a RISC-V based microcontroller. The FFT implementation increases the performance by 20.7% of the Mel-coefficients extraction compared to the software implementation. The implementation occupies a 1987 Look-Up Tables (ALUT) and 244 Flip-Flops (FF), and 16384 Memory Bits in 256 bits configuration, representing a 77% smaller than the implemented RISC-V processor.