针对在生猪音频识别中单一特征参数无法充分地表征猪声信息的问题,提出了基于梅尔倒谱系数(Mel Frequency Cepstral Coefficient,MFCC)与伽马通倒谱系数(Gammatone Frequency Cepstral Coefficient,GF-CC)的生猪音频融合特征生成方法。...针对在生猪音频识别中单一特征参数无法充分地表征猪声信息的问题,提出了基于梅尔倒谱系数(Mel Frequency Cepstral Coefficient,MFCC)与伽马通倒谱系数(Gammatone Frequency Cepstral Coefficient,GF-CC)的生猪音频融合特征生成方法。首先,以5种猪声为研究对象,利用功率谱减法和双门限端点检测法对猪声样本进行预处理。其次,提取MFCC、GFCC和它们的一阶差分参数,将MFCC+∆MFCC、GFCC+∆GFCC直接叠加得到高维的融合特征,为了降低高维特征的冗余度,利用增减分量法对其进行降维,最后将降维后的融合特征输入至Bi-LSTM网络模型进行分类识别。实验结果表明,相对于传统的单一特征MFCC、GFCC在识别率上分别提升了14.33%和18.63%,且在不同噪声环境下,融合特征具有比其他特征更好的鲁棒性和识别性能。展开更多
文章研究了一种融合人工智能的音频信号降噪技术。首先,构建一个包含滤波方法、人工智能等关键组成部分的音频降噪框架。其次,针对音频降噪框架中自适应滤波与长短期记忆(Long Short Term Memory,LSTM)网络相结合的声音降噪方法,详细介...文章研究了一种融合人工智能的音频信号降噪技术。首先,构建一个包含滤波方法、人工智能等关键组成部分的音频降噪框架。其次,针对音频降噪框架中自适应滤波与长短期记忆(Long Short Term Memory,LSTM)网络相结合的声音降噪方法,详细介绍其数学原理和实现过程。最后,采用Noisex-92数据集对所提方法进行全面的测试与评估。结果表明,文章提出的方法在信噪比(Signal to Noise Ratio,SNR)和信号失真比(Signal Distortion Ratio,SDR)上均取得了显著提升。展开更多
This paper describes an audio digital signal-processing toolkit that the authors develop to supplement a lecture course on digital signal processing (DSP) taught at the department of Electrical and Electronics Enginee...This paper describes an audio digital signal-processing toolkit that the authors develop to supplement a lecture course on digital signal processing (DSP) taught at the department of Electrical and Electronics Engineering at the University of Rwanda. In engineering education, laboratory work is a very important component for a holistic learning experience. However, even though today there is an increasing availability of programmable DSP hardware that students can largely benefit from, many poorly endowed universities cannot afford a costly full-fledged DSP laboratory. To help remedy this problem, the authors have developed C#.NET toolkits, which can be used for real-time digital audio signal processing laboratory. These toolkits can be used with any managed languages, like Visual Basic, C#, F# and managed C++. They provide frequently used modules for digital audio processing such as filtering, equalization, spectrum analysis, audio playback, and sound effects. It is anticipated that by creating a flexible and reusable components, students will not only learn fundamentals of DSP but also get an insight into the practicability of what they have learned in the classroom.展开更多
文中介绍了一种语音预处理系统的设计思路以及在FPGA平台下的实现方法。系统由模拟前端电路(Analog Front End,AFE)、高速Flash存储电路和PC端的数据处理模块构成。该系统主要功能是对语音信号实行预处理,通过AFE中的模拟回环功能获得...文中介绍了一种语音预处理系统的设计思路以及在FPGA平台下的实现方法。系统由模拟前端电路(Analog Front End,AFE)、高速Flash存储电路和PC端的数据处理模块构成。该系统主要功能是对语音信号实行预处理,通过AFE中的模拟回环功能获得预处理过的语音模拟信号,同时可向PC端发送未经处理的原始数据,并结合LMS自适应滤波算法可滤除杂波,得到较为纯净的语音数字信号。相比较其他语音处理方式,该系统既可将原始语音信号处理成低噪的数字信号,也可通过内部回环模式得到底噪且增益可放大的模拟信号。展开更多
文摘针对在生猪音频识别中单一特征参数无法充分地表征猪声信息的问题,提出了基于梅尔倒谱系数(Mel Frequency Cepstral Coefficient,MFCC)与伽马通倒谱系数(Gammatone Frequency Cepstral Coefficient,GF-CC)的生猪音频融合特征生成方法。首先,以5种猪声为研究对象,利用功率谱减法和双门限端点检测法对猪声样本进行预处理。其次,提取MFCC、GFCC和它们的一阶差分参数,将MFCC+∆MFCC、GFCC+∆GFCC直接叠加得到高维的融合特征,为了降低高维特征的冗余度,利用增减分量法对其进行降维,最后将降维后的融合特征输入至Bi-LSTM网络模型进行分类识别。实验结果表明,相对于传统的单一特征MFCC、GFCC在识别率上分别提升了14.33%和18.63%,且在不同噪声环境下,融合特征具有比其他特征更好的鲁棒性和识别性能。
文摘文章研究了一种融合人工智能的音频信号降噪技术。首先,构建一个包含滤波方法、人工智能等关键组成部分的音频降噪框架。其次,针对音频降噪框架中自适应滤波与长短期记忆(Long Short Term Memory,LSTM)网络相结合的声音降噪方法,详细介绍其数学原理和实现过程。最后,采用Noisex-92数据集对所提方法进行全面的测试与评估。结果表明,文章提出的方法在信噪比(Signal to Noise Ratio,SNR)和信号失真比(Signal Distortion Ratio,SDR)上均取得了显著提升。
文摘This paper describes an audio digital signal-processing toolkit that the authors develop to supplement a lecture course on digital signal processing (DSP) taught at the department of Electrical and Electronics Engineering at the University of Rwanda. In engineering education, laboratory work is a very important component for a holistic learning experience. However, even though today there is an increasing availability of programmable DSP hardware that students can largely benefit from, many poorly endowed universities cannot afford a costly full-fledged DSP laboratory. To help remedy this problem, the authors have developed C#.NET toolkits, which can be used for real-time digital audio signal processing laboratory. These toolkits can be used with any managed languages, like Visual Basic, C#, F# and managed C++. They provide frequently used modules for digital audio processing such as filtering, equalization, spectrum analysis, audio playback, and sound effects. It is anticipated that by creating a flexible and reusable components, students will not only learn fundamentals of DSP but also get an insight into the practicability of what they have learned in the classroom.
文摘文中介绍了一种语音预处理系统的设计思路以及在FPGA平台下的实现方法。系统由模拟前端电路(Analog Front End,AFE)、高速Flash存储电路和PC端的数据处理模块构成。该系统主要功能是对语音信号实行预处理,通过AFE中的模拟回环功能获得预处理过的语音模拟信号,同时可向PC端发送未经处理的原始数据,并结合LMS自适应滤波算法可滤除杂波,得到较为纯净的语音数字信号。相比较其他语音处理方式,该系统既可将原始语音信号处理成低噪的数字信号,也可通过内部回环模式得到底噪且增益可放大的模拟信号。
