Assessing the effect of sound file compression and background noise on measures of acoustic signal structure
Assessing the effect of sound file compression and background noise on measures of acoustic signal structure
复制标题
DOI：
10.1080/09524622.2017.1396498
复制
发表时间：
2019-01-02
影响因子：
1.8
通讯作者：
Webster, Michael
中科院分区：
文献类型：
--
作者：
Araya-Salas, Marcelo;Smith-Vidaurre, Grace;Webster, Michael
动物声音信号的研究是生态学和进化中许多领域的核心工具，但动物声音记录的分析方法和来源的多样性对进行稳健的声学分析提出了重要挑战。声音文件压缩和背景噪声都可能影响声学分析，尽管很少有人关注它们各自的影响。我们评估了这些因素的影响，通过评估系统偏差（即偏差）和测量误差（即精度），他们产生的光谱参数和两个（不）相似性的方法（动态时间扭曲的频率轮廓和互相关），这是目前最常用的方法用于定量表征的声学信号。对来自不同鸟类的各种信号进行了测量，并将未压缩文件和从使用两种最常见的mp3编码器（Fraunhofer和LAME）生成的mp3编码文件中获得的解压缩文件进行了比较。测量结果还比较了一系列合成产生的背景噪声水平。压缩没有显着偏见的任何声学或相似性测量。然而，表示单个极值（例如峰值频率）的声学参数以及动态时间弯曲距离的精度受到压缩的强烈影响。高背景噪声使大多数与能量分布相关的参数（如谱熵）产生偏差，并影响大多数声学参数和动态时间弯曲的精度。总的来说，压缩和背景噪声确实对声学分析产生了相当大的影响。我们提供建议，以避免潜在的陷阱，并最大限度地提高信息，可以可靠地获得。
The study of animal acoustic signals is a central tool for many fields in ecology and evolution, but the diversity of analytical methods and sources of animal sound recordings poses important challenges for carrying out robust acoustic analyses. Sound file compression and background noise may both affect acoustic analysis, although little attention has been paid to their respective effects. We evaluated the effect of these factors by assessing the systematic deviation (i.e. bias) and measurement error (i.e. precision) that they generate on spectrographic parameters and two (dis)similarity methods (dynamic time warping on frequency contours and cross-correlation), which represent the most common methods currently used for quantitative characterization of acoustic signals. Measurements were taken across a wide range of signals from a diverse group of bird species, and compared between uncompressed files and decompressed files obtained from mp3-encoded files generated using the two most common mp3 encoders (Fraunhofer and LAME). Measurements were also compared across a range of synthetically-generated background noise levels. Compression did not significantly bias any of the acoustic or similarity measurements. However, the precision of acoustic parameters representing single extreme values (e.g. peak frequency) as well as dynamic time warping distances, was strongly affected by compression. High background noise biased most energy distribution-related parameters (e.g. spectral entropy) and affected the precision of most acoustic parameters and dynamic time warping. Overall, compression and background noise did have considerable effects on acoustic analyses. We provide recommendations to avoid potential pitfalls and maximize the information that can be reliably obtained.