基于快速傅里叶变换算法的蜂鸣器音频分析技术

Buzzer audio analysis technique based on fast Fourier transform algorithm

为了深入理解与控制蜂鸣器音效的演奏与生产,研究基于快速傅里叶变换设计了蜂鸣器的音频分析算法,利用快速傅里叶变换与改进的LeNet-5网络实现了声谱特征图的识别分类。仿真实验表明,经关系系数、欧氏距离和余弦距离指标的验证,研究改进的傅里叶-卷积网络模型在现场可编程门阵列上的计算结果与电脑计算结果吻合度较高,计算用时减少55.97%。与其他音频识别算法相比,研究设计的算法在平均绝对值误差与均方根误差上表现较好,取值最小分别为0.587、0.561,损失效果较好,与真实结果的吻合程度较高。此次研究基于快速傅里叶变换设计的蜂鸣器音频分析算法能够有效实现音频事件的识别与分类,对于实现蜂鸣器音乐演奏的交互具有实用价值。

In order to deeply understand and control the performance and production of buzzer sound effect,the research designed the audio analysis algorithm of buzzer based on the fast Fourier change,and firstly,the fast Fourier change and the improved LeNet-5 network were used to achieve the recognition and classification of the sound spectrum feature map.Simulation experiments show that,verified by the relationship coefficients,Euclidean distance and cosine distance indicators,the research improved Fourier-convolutional network model on the field-programmable gate array has a high degree of agreement with the results of the computer calculations,and the computation time is reduced by 55.97%.Compared with other audio recognition algorithms,the research-designed algorithm performs better in the average absolute value error and root mean square error,taking the smallest value of 0.587 and 0.561,respectively,with a better loss effect and a better match with the real results.The buzzer audio analysis algorithm designed in this research based on fast Fourier variation can effectively achieve the identification and classification of audio events,which is of practical value for achieving the interaction of buzzer music playing.