MobySound is made up mainly of sound sets. A sound set contains, at minimum, sound files, a README file, and annotation files.

Sound files: These are recordings of the calls of a given species. Each sound file may have sounds from one or more individuals. Some of the sound files may contain only noise and interfering sounds, which can be useful in training an automatic call recognition algorithm to distinguish calls of interest from other sounds. Sound files are in WAVE (.wav) or AIFF (.aif) format.

Readme file: This provides information about the sound and annotation files: (1) Where and when the sounds were recorded. (2) The recording equipment used. (3) Why the recordist believes the calls in the sound files came from the given species. (4) Whether calls from more than one individual are present in some of the sound files. (5) Whether calls from separate individuals have separate annotation files, and if so, how the recordist determined which call came from which individual. (6) Any other information that might be pertinent to the sound set.

Annotation files: Each annotation file is associated with a particular sound file. It describes where in the sound file the calls of interest occur. This description, for each call, specifies the time and frequency bounds of each call. Time bounds are specified as start- and end-times of the call, in seconds since the start of the file. Frequency bounds are specified as low and high frequencies, in Hertz. Each call description also contains the signal-to-noise ratio (SNR) of the call in decibels, as measured by the procedure described in this paper. If calls of separate individuals are distinguished in this sound set (as described in the README file), then each individual's calls have a separate annotation file. Thus each sound file might have several annotation files.

About MobySound