Laboratory of Biological Modeling, The Rockefeller University, 1230 York Avenue, New York, NY 10065, USA.
Comput Intell Neurosci. 2012;2012:261010. doi: 10.1155/2012/261010. Epub 2012 Jun 4.
The singing of juvenile songbirds is highly variable and not well stereotyped, a feature that makes it difficult to analyze with existing computational techniques. We present here a method suitable for analyzing such vocalizations, windowed spectral pattern recognition (WSPR). Rather than performing pairwise sample comparisons, WSPR measures the typicality of a sample against a large sample set. We also illustrate how WSPR can be used to perform a variety of tasks, such as sample classification, song ontogeny measurement, and song variability measurement. Finally, we present a novel measure, based on WSPR, for quantifying the apparent complexity of a bird's singing.
幼鸟的歌声变化多样且不规范，这一特征使得现有的计算技术难以对其进行分析。我们在此提出了一种适用于分析此类叫声的方法，即窗口频谱模式识别（WSPR）。WSPR 不是进行成对样本比较，而是衡量样本与大样本集的典型性。我们还说明了如何使用 WSPR 执行各种任务，例如样本分类、歌曲发生测量和歌曲可变性测量。最后，我们提出了一种新的基于 WSPR 的度量方法，用于量化鸟类歌唱的明显复杂性。
