基于深度学习网络的AI拟声检测系统的设计

打开文本图片集

中图分类号：TP391.4 文献标识码：A 文章编号：2096-4706（2025）22-0035-05

Design of Al Onomatopoeia Detection System Based on Deep Learning Network

LAN Jiayun， DUAN Xiaoxia， WANG Liying （Guangzhou College of Technology and Business，Guangzhou 51o850， China）

Abstract： In theeraof rapid development of information technology，AIonomatopoeia technology has brought changes to various industries，butitasalsoaisedsafetyandethical challenges，especiallywhen itisdicult todistingushbeween real human voices andAI-generated voices.The purposeofthispaper is todesignand implementan effcient AIonomatopoeia detectionsystemto distinguishAIonomatopoeia fromreal human voices.Thesystem uses ahybrid modelmethodcombining RecurrentNeuralNetwork（RNN）and VariationalAutoencoder（VAE）tolearn thedeprepresentationofsoundfeaturesand capture the diffrencesbetweenAIonomatopoeiaandreal human voices.The sound is comprehensivelyanalyzed fromthree dimensions of spectralcharacteristics，prosodic characteristicsand sound quality.Then，the analysisofemotional speech synthesistechnologyisintroducedtoenhance themodel'sabilitytorecognizeAIonomatopoeia.Experimentsshowthatthe detection system performs wellon multiple public datasets， which proves its effectiveness and feasibility.

Keywords：AI onomatopoeia; speech synthesis; Recurrent Neural Networks; variational autoencoder; affective speech synthesis;sound analysis

0 引言

随着科技的飞速发展，特别是在生物信息学领域，基因表达数据的积累已达到空前规模，这既展示了生物学研究的广阔前景，也带来了数据分析的新挑战。（剩余6972字）