智能家电声纹识别测试用声源选取的研究

Research on Sound Source Selection for Smart Home Appliance’s Voiceprint Recognition Test

声纹识别技术早被广泛应用于刑侦、支付方面,近几年随着智能家电的深入发展,该技术也开始被智能家电厂商重视并应用。但在测试智能家电的声纹识别能力时,行业内在测试用声源的选取方面产生了较大分歧,主要集中在真人声源和标准库的合成声源。本文通过matlab基于BP神经网络模拟声纹识别智能家电系统,并通过用两种声源进行测试,对比试验结果的离散程度,得出用软件合成的标准语音库声源进行试验更可靠的结论。

Voiceprint recognition technology has been widely used in criminal investigation and payment. In recent years, with the development of smart appliances, this technology has also been applied by smart appliances manufacturers.However, when testing the ability of sound fingerprint identification of smart home appliances, there is a big divergence in the selection of sound source for testing, mainly focusing on the real sound source and the standard library of synthetic sound source. Based on BP neural network simulation voiceprint recognition intelligent home appliances system of Matlab, the paper uses two kinds of sound source tests to compare test results of the degree of dispersion. It is concluded that it is more reliable to use the standard speech library synthesized by software to carry out the experiment.