语音增强算法研究系列笔记 – 语音噪声分类及特点
噪声特征及分类
噪声是指干扰正常信号特性的所有信号，即信号中不必要的特征。
[En]
Noise refers to all signals that interfere with the characteristics of a normal signal, that is, unnecessary features in the signal.
根据噪声对语音频谱干扰方式1的不同，可以将噪声分为加性噪声和乘性噪声；
加性噪声
加性噪声是指当噪声对语音的干扰在时间域中由两个信号相加表示时。
[En]
Additive noise means that when the interference of noise to speech is represented by the addition of the two signals in time domain.
很明显，噪声和语音在频域中是相加的。
[En]
It is obvious that noise and speech are additive in the frequency domain.
实际环境中背景噪声可以看成加性噪声，如风扇的声音、汽车引擎声、周围人说话声等；加性噪声是对背景噪声一种比较贴切的表述；麦克风等声音采集设备在正常工作的范围内，可以近似看成一个线性系统，即产生信号的幅度和声强呈正比； 从能量角度 1看背景噪声和语音的声强是叠加关系，两者对麦克风共同作用形成的带噪语音信号等于各信号之和；目前针对这类噪声的研究最为常见
在语音增强系统中，常用的加性噪声有高斯白噪声、粉红噪声和工厂噪声2；其中高斯白噪声和粉红噪声属于平稳噪声；而工厂噪声属于非平稳噪声，存在一段尖锐的类似脉冲噪声的噪声
白噪声
白噪声是指功率谱密度在整个频域内均匀分布的噪声。在所有频率上具有相同能量的随机噪声称为白噪声。
[En]
White noise refers to the noise in which the power spectral density is uniformly distributed in the whole frequency domain. Random noise with the same energy at all frequencies is called white noise.
; 粉红噪声
粉红噪声定义为在与频带中心频率成正比的带宽（如倍频程带宽）内具有相等功率的噪声或振动；粉红噪声的频率分量功率主要集中在中低频段，从波形角度上看，粉红噪声是分形的，在一定的范围内音频数据具有相同或类似的能量，从功率（能量）角度来看，粉红噪声的能量从低频向高频不断衰减，曲线为1/f，通常在线性频率坐标下，其功率谱密度以3dB每频倍程的速率下降；
工厂噪声
工厂噪声一般是指工业设备和运行过程中产生的噪声。
[En]
Factory noise generally refers to industrial equipment and the noise generated during operation.
乘性噪声
乘性噪声是指噪声和语音在频域上的乘法关系，以及在时间域和语音上的卷积关系，所以又称为卷积噪声；在实际应用中，乘性噪声主要体现在语音采集和麦克风传输中电话信道和无线信道的选频特性；乘性噪声可以通过状态滤波等变换转化为加性噪声。
[En]
Multiplicative noise refers to the multiplicative relationship between noise and speech in frequency domain, and convolution in time domain and speech, so it is also called convolution noise; in practical application, multiplicative noise is mainly reflected in the frequency selection characteristics of telephone channel and wireless channel in speech acquisition and microphone transmission; multiplicative noise can be transformed into additive noise by some transformation like state filtering.
根据噪声随时间变化的统计特性，噪声可分为周期噪声、脉冲噪声、缓变噪声和平稳噪声。
[En]
According to the varying degree of noise statistical characteristics with time, noise can be divided into periodic noise, impulse noise, slowly varying noise and stationary noise.
周期噪声
发动机和城市功率产生的干扰是周期性噪声，在频域上具有多个离散线谱的特点，可以用梳状滤波器进行滤波，然后进行数字信号处理。实际环境中产生的周期性噪声不仅包含简单的线性谱分量，而且是由许多窄带频谱组成的。这种类型的噪声通常是时变的，并且与语音信号的频谱重叠，通常需要自适应滤波来自动识别和区分噪声。
[En]
The interference produced by the engine and the city power are periodic noise, which is characterized by many discrete line spectra in the frequency domain, which can be filtered by comb filters and processed by digital signal processing. The periodic noise generated in the actual environment does not simply contain only linear spectral components, but is composed of many narrowband spectra. This type of noise is often time-varying and overlaps with the frequency spectrum of the speech signal, and adaptive filtering is often needed to automatically identify and distinguish the noise.
脉冲噪声
脉冲噪声是由点火和放电产生的，在时域波形上表现为一个窄脉冲，只要脉冲噪声不是太密集，就可以通过内插去除。
[En]
Impulse noise is produced by ignition and discharge, which is shown as a narrow pulse in the time domain waveform, which can be removed by interpolation as long as the impulse noise is not too dense.
缓变噪声
慢变噪声是实际情况中经常遇到的一种噪声，这种噪声的统计特性会随着时间的推移而缓慢变化，而人群噪声是一种典型的慢变噪声。
[En]
Slowly varying noise is a kind of noise often encountered in practical situations, and the statistical characteristics of this kind of noise will change slowly with time, and crowd noise is a typical slowly varying noise.
平稳噪声
平稳噪声是指噪声的统计特性不随时间变化；由于噪声源的复杂性，日常生活中遇到的大多数噪声都是非平稳的，但对平稳噪声的研究是噪声分析的基础。
[En]
Stationary noise means that the statistical characteristics of noise do not change with time; because of the complexity of noise sources, most of the noise encountered in daily life is non-stationary, but the study of stationary noise is the basis of noise analysis.
根据噪声覆盖的频率范围，噪声可分为全频带噪声(又称宽带噪声和窄带噪声)。
[En]
According to the frequency range of noise coverage, noise can be divided into full-band noise (also known as broadband noise and narrowband noise).
全频带噪声
覆盖了信号全部频率带的噪声称为全频带噪声或宽带噪声；其来源有很多，如热噪声、气流（如风）、呼吸噪声、量化噪声以及各种随机噪声源；对于平稳的全频带噪声可认为是高斯白噪声；对于不具有白色频谱的噪声，可以先进行白化处理3，然后转化为白噪声；
窄带噪声
仅覆盖信号部分频段的噪声称为窄带噪声，也称为频带选择性噪声；口哨噪声是窄带噪声的一种。
[En]
The noise that covers only part of the frequency band of the signal is called narrowband noise, also known as band selective noise; “whistle” noise is a kind of narrowband noise.
其他噪声
此外，根据不同学科研究的噪声类型，有许多噪声分类。我不会在这里重复它们。请详细了解相关领域的噪声分类。
[En]
In addition, there are many noise classifications according to the types of noise studied in different disciplines. I will not repeat them here. Please learn about the noise classification in related fields for details.
Original: https://blog.csdn.net/weixin_44985601/article/details/108818934
Author: Light_Laser
Title: 语音增强算法研究系列笔记 – 语音噪声分类及特点
原创文章受到原创版权保护。转载请注明出处：https://www.johngo689.com/526331/
转载文章受原作者版权保护。转载请注明原作者出处！