Arbimon Acoustics Banner



Recordings and Analytical Tools

        Automated recording devices have improved our ability to simultaneously monitor biodiversity in many sites.  This has resulted in millions of audio recordings, but it is often difficult to extract useful data from these recordings.  To help solve this problem ARBIMON-acoustic provides a database and analytical tools that integrate file management, audio processing, sonogram evaluation, species-specific identification modeling and verification in a web-based application.

 

The components that make up this interface include:

 

1. Visualizer - This module is used for viewing, listening, and annotating recordings.  Presently, we have >1,000,000 recordings that you can listen to.  The interface can accept recordings of any length and from most recording devices. 

 

2.  Species validation – This tool allows the user to specify which species/call is present or absent in each recording.  In addition, the user can determine if the particular call is correctly marked by the automated ROI (regions of interest) generator. 

 

 

3. Model builder – This component has four steps. 

a. Training data – the first step in developing a species-specific model is to provide training data for the model by identifying examples of the song/call that the user wants to model.  The user describes a call sequence by selecting a series of ROIs from a recording.  This process is repeated with other examples of the call to provide the program with additional training samples.

 

b. Model creation – the next step is to define the model (number of notes, and the length of the sequence) and train the model.

 

c. Applying model – The initial model then can be applied to any group of recordings in the database.  Usually, the user will apply the model to batches of 500 recording, looking for matches that can be incorporated into the training data of the model.  Once the user is satisfied with the model it can then be tested against the validation data.

 

d. Validation – In this step, the system applies the model only on the recordings where the user has validated the presence/absence of the species. In this step the user is provided with an error matrix and statistics on the accuracy, precision, and kappa. Based on these statistics the user can modify the model by varying the range of values (e.g. minimum frequency, duration) used in determining which ROIs are used in the model. 

 

4. Apply model, data export, and publish model – In this step, the user can apply the model to their complete data set.  The results from this analysis can be exported to excel for further analyses.  In addition, the user can “publish” the model, making it available to other users and other projects.