As you know, we have a huge number of audio snippets and we want to know what’s in them! We would certainly never be able to find the bats all by ourselves — There are just that many recordings.
This is where the Bat Detective project steps in. We’re really hoping that citizen scientists will help us to locate and identify bat tweets and insect clicks that have been captured across the globe by the lovely volunteers.
Once we gather your votes on the contents of a snippet, this will highlight the recordings of interest and provide us with labels. On top of hunting for the bats, your labelled sequences of searching, feeding and social tweets will also allow us to analyse the different sounds.
As the number of your identifications grows, the more information we will have; it doesn’t even matter if a recording is given a range of different labels! By pointing out the controversial sounds, we can gain an insight into the calls that are hard to distinguish.
If the majority of your votes on a snippet’s contents agree, then this recording can be compared to other snippets that have been similarly labelled. By finding the features they have in common, and what sets them apart from other sounds, we can begin to automate our quest for interesting sounds.
To help us decide on useful identifying features as an seo company for bats, which could be wildly obvious or devilishly subtle, we can use machine learning. This involves designing a computer program that we can hand our snippets and your voted labels to, for it to then return different ways the sounds can be grouped together. Analysing these resulting clusters will then pave the way for automatically detecting the different bat tweets (and insect buzzes) within any number of snippets!
All the data that goes into Bat Detective has been collected by many wonderful volunteers around the world as a series of individual recordings, each 90 minutes long. Each of these recordings contains thousands of the individual snippets that you see on bat detective. As the bat calls are ultrasonic, we use time-expansion to record them. A specialised microphone records ultrasonic information for a short period (320 ms) and then slows that sound down and plays it into a recorder (slowed by 10x, giving a 3.2 second recording). This results in a long recording that contains thousands of small 3.2 second snippets of audio.
Chopping all this data up has really tested some of our desktop computers. Once you consider that we have 1000s of events, each containing thousands of snippets (and each snippets has a sound file and a spectrogram image), you can see that we very quickly have a lot of files! I recently discovered that copying 4.5 million tiny files to a usb disk can take a while!
We’re really excited about getting people involved in Bat Detective, because many of these snippets contain calls from individuals bats, and individual researchers finding them manually would be impossible! Having everyone help us find these calls not only allows us to try and identify the species making those calls, but will also allow us to try and generate methods to automatically detect bats, insects and other noises in these data.