Song Scope


Song Scope software is a powerful tool to efficiently review recordings made by Song Meter or other recording equipment. At the core of Song Scope is a spectrogram viewer that let’s you scroll through large audio files (up to 2GB) to visually inspect spectrograms. With practice, you will recognize the visual patterns corresponding to the vocalizations of interest. Viewing audio files as spectrograms is typically 10-20 times more time efficient than listening to the same recordings with your ears. Of course, you can always play selected samples of the recordings in Song Scope to let your ears help with identification.

Song Scope also has powerful patented classification algorithms designed to automatically scan through many recordings to look for specific patterns of interest. This is very useful for finding “needles in haystacks”. You can think of the Song Scope classifiers (called “Recognizers”) as a sieve for sifting through thousands of hours of audio recordings to find the most likely occurrences of a specific vocalization of interest. The results can then be quickly reviewed and confirmed.

Free Evaluation

We encourage all potential Song Scope customers to trial Song Scope for 15 days for free. You can request your free trial here.

Building Classifiers in three steps:

Annotate Known Vocalization

Generate a “Recognizer”

Search Field Recordings with One Click
Start with your own recordings of the vocalizations you are interested in studying. This is the “training data” used to teach Song Scope how to recognize a new vocalization. With Song Scope, you can quickly listen to portions of these recordings and label them with “annotations”. In this example, the song of a Cerulean Warbler is selected and annotated to be used as training data. Next, Song Scope builds a “recognizer” by analyzing the annotated recordings you provide. In this example, a recognizer is constructed from training data containing a number of known samples of Cerulean Warbler vocalizations. These vocalizations are “imported” into Song Scope, and you can adjust a handful of parameters to guide Song Scope’s algorithms for better results. At the push of the “Generate Recognizer” button, Song Scope automatically analyzes the selected vocalizations to build a “recognizer” using Song Scope’s patented algorithms. Finally, you can use the “recognizer” to quickly scan new recordings to detect similar vocalizations by pressing the “Start Scan” button. In this example, several instances of possible Cerulean Warbler vocalizations are listed in a spreadsheet. By clicking on one of the rows, Song Scope automatically views the spectrogram of the detected vocalization. With sufficient training data, Song Scope is typically 80% accurate for even complex and highly variable vocalizations in even somewhat noisy environments. However, results may vary depending on a number of factors.

What to expect:

  • Building a good classifier requires good training data, patience, and a lot of trial and error. There are many different settings for adjusting classification parameters and it may take some amount of experimentation to find the optimum settings for any given situation.
  • There is no silver bullet. The results will include true positives (correct detections), false positives (misclassifications), and false negatives (missed detections). Reducing the false negative rate will often lead to increasing the false positive rate.
  • Mileage will vary. For some situations, depending on the variability of the vocalization and competing sounds in your environment, the classifier might be invaluable to your work. In other situations, it may not. The only way to know is to try.
  • Even classifiers with high error rates can help reduce your workload. By reducing large data sets to just the likely matches, even with a high number of false positives, you can save time by quickly reviewing the results rather than scrolling through spectrograms manually.

Create a “Song Scape” from your Song Meter recordings


Song Scape is a revolutionary new capability fully integrated with Song Scope that lets you visualize sounds on large temporal scales. This is like a spectrogram, but with the horizontal axis measuring months and years instead of seconds. The animation to the left is a Song Scape created from 3,530 individual recordings made hourly through the spring and summer of 2009 at an undisclosed site. Each frame in the animation represents the sound energy present at different times of the day. You can clearly see two species of frogs calling in the spring at night and the dawn chorus in the morning.

By double clicking on a pixel in the Song Scape plot, Song Scope automatically opens the corresponding source recording so you can zoom in to the actual individual species vocalizations. Song Scape data can also be exported for analysis by other statistical programs.

A Song Scape is represented by a “Song Scape Summary” file, or .sss file. The .sss file is first created with a set of parameters such as the desired frequency and temporal resolution. Next, recordings can be added to the .sss file in batches. In this step, Song Scope scans the recordings to calculate spectral parameters, automatically splitting or combining recordings to meet the temporal resolution specified. Finally, the Song Scape can be opened, manipulated, and viewed, while new recordings can be added to the Song Scape as they continue to be collected from the field.

How does it work?

For more information about the algorithms and their performance, please read our white paper: Automatically Identifying Animal Species from their Vocalizations.

You may also be interested in reading our article featuring Song Scope software in a demonstration deployment for the U.S. Forest Service San Dimas Technology and Development Center in cooperation with the Cornell Lab of Ornithology Bioacoustics Research Program: “Automatic Detection of Cerulean Warblers Using Autonomous Recording Units and Song Scope Bioacoustics Software”.

Song Scope User’s Manual may be found on our Documentation page.