Research And Design On Reconfigurable Audio Analysis System

The audio signal analysis system is the carrying platform for audio processing.PYNQ-FPGA combines ARM with FPGA and provides a Python development environment.Compared with traditional FPGA,its Overlay structure can reconfigure resources at any time,providing a reconfigurable experimental platform for audio analysis.This paper studies the audio analysis technology based on PYNQ-FPGA,aiming to design a reconfigurable audio analysis system,which can efficiently perform spectrum analysis and feature recognition of audio signals.During operation,the system will reconstruct resources according to different functions,thus providing the reconfigurability of functions and the reconfigurability of resources,which can save resources to the maximum extent and improve operating efficiency.Firstly,the technical foundation of audio analysis is studied.The core of audio signal analysis is spectrum analysis.Since the audio signal is a non-stationary signal,the typical method of frequency domain conversion is short-time Fourier transform.That is dividing a signal into frames and performing Fourier transform on each frame of signal separately.After obtaining the spectrum,The Mel frequency cepstral coefficients are extracted and used as audio features for further analysis.Then,the analysis and recognition method of audio features are studied.The convolutional neural network is selected as the realization method,its basic structure is demonstrated.On the basis of completing the research of audio analysis technology,the overall structure of the reconfigurable audio analysis system based on PYNQ-FPGA is given.The structure includes two parts: hardware platform and software module.In the hardware platform design,the chip selection of FPGA is completed first,and the component selection and hardware circuit of its related peripherals and interface circuits are described.Then the audio codec chip WM8731 is researched and designed,and its basic circuit,function,structure and communication protocol are analyzed.Based on the hardware platform,the software design implements three modules: audio acquisition and generation,audio analysis and result display.The IP sub-modules used by each module are introduced,and drivers of each IP are given.For the audio analysis module,the implementation of short-time Fourier transforms and Mel filter cepstral coefficient extraction was introduced,and Convolutional neural network was built to realize the recognition of "up","down","left" and "right" in Chinese audio,and the It is ported to PYNQ-FPGA.Then the operation flowchart of each module is given.Finally,the system is tested and analyzed.The data path is tested first,and then the system test is performed.The test results verify the functional integrity and the reconfigurability of the system.