摘要
为全面描述电子琴声音信号的特性,增强系统的鲁棒性和泛化能力,设计基于多特征融合与迁移学习的电子琴声音信号自动化识别系统。设计声音信号采集与处理模块,实时采集声音信号,并经由频谱分析,得到声音信号的短时傅里叶变换、常数Q变换和对数梅尔频谱图;在上位机中采用TL-VGG16迁移学习的特征提取网络,分别在频谱图中提取声音信号的振幅、音高与音色特征,增强系统的鲁棒性和泛化能力;对特征进行多特征融合,经由全连接层输出声音信号自动化识别结果。实验证明,该系统的识别准确率在0.75以上且相对稳定,且该系统可有效将采集的声音信号转换成频谱图,并提取声音信号特征,完成多特征融合,进而精准自动化识别声音信号。
To comprehensively describe the characteristics of electronic piano sound signals and enhance the robustness and generalization ability of the system,an automated recognition system for electronic piano sound signals based on multi feature fusion and transfer learning is designed.An electronic piano sound signal acquisition and processing module is designed to collect real⁃time sound signals and obtain short⁃time Fourier transform,constant Q transform,and logarithmic Mel spectrogram of the sound signals through spectral analysis;the TL-VGG16 transfer learning feature extraction network in the upper computer extracts amplitude,pitch,and timbre features of sound signals from the spectrogram,enhancing the robustness and generalization ability of the system;perform multi feature fusion on features and output automated recognition results of sound signals through a fully connected layer.Experimental results have shown that the recognition accuracy of the system is above 0.75 and relatively stable.The system can effectively collect sound signals,convert them into spectrograms,extract sound signal features,complete multi⁃feature fusion,and accurately automate the recognition of sound signals.
作者
刘志方
张丽珍
LIU Zhifang;ZHANG Lizhen(North University of China,Taiyuan 030051,China)
出处
《电子设计工程》
2025年第23期177-181,共5页
Electronic Design Engineering
关键词
多特征融合
迁移学习
声音信号
自动化识别
频谱图
multi⁃feature fusion
transfer learning
sound signal
automated identification
spectrogram
