Creating and applying vocalization recognizers to audio files

Byrne, Michael; Smrekar, Briana

This Standard Operating Procedure (SOP) describes the steps involved with developing and applying recognizers to audio files in a Song Scope software environment. All field personnel must read the Song Scope User’s Manual prior to operation, which provides an excellent background of sound properties, the software interface (i.e., menus, navigation) and detailed instructions on the recognizer development and application process. It is also recommended that all personnel implementing this SOP read the forums on Song Scope software located on the Wildlife Acoustics, Inc. website (http://www.wildlifeacoustics.com/). This document, the user manual, and the forum are expected to be used together during the recognizer-development process. Recognizers are species-specific generalized computer algorithms derived from the visual representation (i.e., spectral characteristics or spectrogram) of a vocalization that, in a simplified description, facilitate pattern recognition. Recognizers are applied to audio files in a software environment,—Song Scope Bioacoustics Software in this case— and provide an automated analysis of species occurrences. Recognizers are species specific, and generated from multiple individual vocalizations. It is currently not possible to differentiate among individuals in an audio recording; therefore, all detections from one recording event are considered one individual. It is very important to note that creating effective recognizers is not a “plug-and-chug” exercise. It takes a great deal of creativity, intuition, common sense, and most importantly, patience. Knowledge of the frequency ranges (Appendix A), and spectral and temporal characteristics of all species likely to occur is useful and aids in determining which species have the highest likelihood of generating an effective recognizer. This knowledge base will minimize the trialand- error component of the initial attempts at recognizer development, and assists in determining which components of the vocalization are used for the recognizer. Further, a good background can aid in determining which species’ vocalizations are similar to an extent where they can, or must be pooled, for reliable detection.

Creating and applying vocalization recognizers to audio files

Collections