Overview
Terra users a proprietary set of algorithms to identify bird sounds. Here's the basic process:
1. A sound is detected by a Terra device in the field. The Terra automatically determines if it might be a bird sound, and clips out that section for analysis.
2. The clip is then sent to the Terra ID engine, which analyzes it based on our own proprietary set of algorithms.
3. If the engine IDs the clip as a bird, it saves the ID and the confidence level (how sure Terra is that it's correct, expressed as a percentage), and shares that ID back with the Terra account that submitted the clip.
About Detecting Sounds
Range
It would be nice if we could give an exact distance or sound level that Terras can identify birds from, but that's not easily done since there are a number of factors involved. As a general rule, Terra can hear everything that a person with very good hearing can, plus a little bit more. So if you can hear a faint bird sound standing next to Terra, then Terra can hear it as well. Sound travels with different amounts of efficiency depending on things like temperature, humidity, wind direction and speed, and objects between the sound and the hearer. So you may find that a sound is easier to detect one day and harder the next.
Optimizing Detections
There are a few things you can do to optimize your bird detections with Terra. Remember that there are two factors in getting a good sound for identification: 1. the volume of the sound of the bird 2. the volume of the other sounds around it. In audio recording people refer to the signal-to-noise ratio, which is sometimes used to describe the ability of a microphone to detect a sound without introducing its own, self-made noise (like the hiss you get when you turn up a speaker). It can also be useful to think of the signal-to-noise ratio as the bird sound vs the background sound. For example, if I have a bird making a barely audible "cheep", but the background is silent, then Terra will likely be able to identify it. But if that sound is surrounded by road noise, then the road noise can drown out the cheep and it won't be picked out by the Terra. This can be true even for loud sounds when the background noise is high enough. So overall we want to get as close to bird sounds as possible, but simultaneously avoid any loud background sounds.
That might mean that instead of placing your Terra right under the bird feeder in your front yard, which might pick up a lot of road noise, you might get better sound by placing it around the side of the house, where it can still hear the birds but where the road noise will be largely blocked.
This is also why it's not necessarily the case that placing the device high up will give you the best listening experience. It is true that it may be able to pick up migration sounds and radio tags a bit more easily, but placing the microphone high will also pick up road noise and other sounds well, too. So the benefits may not outweigh the drawbacks ... the additional background sounds could obscure some of the bird sounds, and it also might not be as pleasant a listening experience.
Placing your Terra lower to the ground and close to cover, feeders or a water source will often pick up more birds on the ground, and still pick up migration sounds. It will also help to block extraneous noise like traffic, which is (unfortunately) nearly ubiquitous in most areas where people live. In fact, you might find that putting your Terra on the other side of the house from the road, or behind a small hill or even a mound of dirt can help block those kinds of sounds, and give you more wildlife sounds instead.
Terra's ID Algorithm
How it Learns
The first half of identifying a sound is getting a good recording, which we discussed above. The next part comes down to the Terra ID Algorithm, a proprietary model that we've created to automatically identify bird sounds from those recordings. Our ID engine is not a "one-and-done" creation - it is a dynamic tool that will improve over time. There are a couple of ways we do that: 1. We use the data we collect from the Terra network to grow our training model 2. We work on the algorithms we have and look for areas we can improve 3. We use user feedback to vet our model and further refine it -- see the
Terra ID Review tool.
Over time we will also be adding non-bird identifications, including frogs, insects and mammals.
Bird ID Issues
Some issues you might see with bird ID are:
1. Not identifying birds that you hear clearly. This is something that we will work on in the algorithm - if you have specific sounds that aren't being identified that you think should be, let us know. PLEASE NOTE: Sometimes your IDs will show up sporadically or not at all in the app, which might be due to an outage or to a login issue, not an ID engine issue. If you aren't seeing IDs, please go to your computer and login at
www.terralistens.com/login
to see if the IDs show up on the computer. You can also go to
map.terralistens.com and find your device, click it, and see if it's showing current IDs. If you see good IDs in either place but not the app, then try logging out of the app and logging back in again. If that doesn't work, try deleting the app, reinstalling it, then logging in to the app.
2. IDing birds that aren't there. We've seen a set of birds that are "ghost IDs" - in other words, the Terra thinks it hears a bird when it doesn't. This can happen for one of two reasons: it can be because of silence being amplified too much by the engine, and then trying to find an ID out of that silence; or it can be a dog or siren or other sound that is being mis-IDed. In any case, we are working to remove these common mis-IDs in the engine. Some examples of common "ghost IDs" are Great Horned Owl, Mallard and other ducks (especially at night), Sanderling, Red Crossbill, Common Loon, Common Raven, and Tundra Swan.
Overall, the ID engine works well for the majority of people with the majority of birds, but we want to make it as accurate as it can be, so we'll continue to improve it for as long as we need to, and we thank you for your efforts to help us make Terra everything it can be!