[ad_1]
Recently I was just walking around and cutting down the trees for a woodcut park in Brooklyn with my iPhone in hand. Birds were singing everywhere, but in the evening, I was recording a strange song: This was a steady, metal whistle of Bicknell. Although it is a very colorful, spotless bird, this rare brown bird is a favorite of New York bird owners – but finding it is difficult to find. Unless you hold it in your hand, you will not be able to detect it accurately in terms of its appearance alone, and its lyrics are slightly different from the doppelganger, the most common gray thrush.
I left the trail and just wrote down the artwork, which was filled with background noise and the sound of other birds. But when I put the file on Merlin Bird ID’s new Sound ID app, it listed all the recorded birds, including cardinals and warblers, and was able to distinguish between Bicknell’s weak whistle and those with gray cheeks that were all on record.
Many programs try to identify birds from images and sounds, in a variety of ways — one program I was asked to watch called the drawing of any northern bird, a bird that imitates other birds. But bird breeders and citizen scientists have long relied on the Cornell Lab of Ornithology’s Merlin Bird ID as a way to help detect bird pictures. When I realized they were expanding their operations to birdong, I was quick to try this and was eager to learn more about what makes machines recognizable.
Birdwatchers know about birds and their unique songs, but doing so can be challenging and time-consuming. This is the goal of Merlin Bird ID-assisting those who are still trying to identify things. “The great thing about Merlin is that it’s a non-judgmental friend who can tell you that you’ve been hearing the music ring for 300 years, and he’ll tell you as excitedly as he did the first time,” said Drew Weber, Merlin Bird ID’s project manager.
G / O Media can be sent
I took the program for a test drive last week at Prospect Park in Brooklyn to make sure his success in the original film was not insignificant. Although the city’s habitat makes it an ideal place for bird watching in the summer and fall, with only a handful of birds remaining in the parks during the summer, so the program has the opportunity to feature well-known birds.
I stood on a tree near the doorstep of the southwest, where the people of Baltimore were singing from the pine tree. I started the Sound ID app, hit the record, and kept my phone in my head. The program showed me the program – a frequent form that was filmed over time – and he immediately said “Robin American” yes, the robin had started singing behind me. I tried again, and this time, the house sparrow began to sweep. The program showed me a picture of a house sparrow. I tried one last time, and as the oriole singers sang, the quick toilet settled down; the app responded that it also ignored the drawings in order to identify something else. I think this shows the curiosity that this app can detect, but I was disappointed that it failed to recognize an oriole – a common bird – in this simple period of time.
Entering the park’s forest, I would open the app and record any other birds I might encounter. It identified well “pew-pew-pew ”music, even as the cardinal began to make a high-stakes ticket, the program cheerfully declared that I now listen to a large bird, a fish-eating snail. The loud, high-pitched “seeee” sound of the cedar was seen strangely in the game, though the voice was not recognized, and instead a video of a fighting fight appeared as one began to sing in the distance (a song I heard as “a drunken man trying to make sense”).
Merlin’s ID overwhelmed me, though; I have never heard of several distant records, and at the same time the program gave the idea of a shooting for Acadian, a bird southeast of the forests that are not uncommon in New York but are sometimes nested in Prospect Park. I walked into the woods, because the program heard the bird better than I did. Apparently, soon I was standing under a tree when the little green bird sang loudly “pwee-tseet”
Merlin Bird ID is not just a popular app, however; and the results of thousands of bird watchers and their fellow scientists who have donated more than a million songs to the Cornell Macaulay Library through the eBird program over the past few years. Given the amount of data, Weber and Macaulay Library researcher Grant Van Horn, as well as other members of the Cornell Lab of Ornithology, wondered last summer what it would take to create a Merlin Bird ID identification program.
Voice recognition is a problem for image recognition, Van Horn explained. Researchers at Caltech and Cornell Tech had already developed a tool for recognizing bird pictures that use images from the Macaulay Library to create a Merlin Photo ID. Sound ID converts sound into spectrogram images, and converts them, and then computer-assisted visual equipment compares these scenes with existing bird drawings.
The key to identification is a strong training depot – which required the help of citizen scientists, explained Weber. Like my Bicknell painting, Macaulay Library art often has a wide variety of background music. A team of volunteer translators participated in training from more than 400 species of North American birds, drawing boxes around and recording each type of bird. The result was stored with about 250,000 words, each box equal to one color only. Users of the app can either download the file or record the birds, and the app returns any hearing bird to every three seconds of listening. The team also taught the changes in a variety of background techniques, including the amount of Google Audioetet, so that the app could hear how non-birds sound.
There are other well-known birdong species, most notably the Cornell Lab of Ornithology, and Chemnitz University of Technology, which also runs the BirdNET Sound ID program. However, the software has a different purpose: BirdNET serves as a scientific research tool, while Merlin is a science-based bird information program that also includes image and Q + Identification, field guide, and data from the citizen’s eBird observer database birds, words, and pictures. Information from eBird also helps to introduce Merlin Sound and Photo ID formats; they rely on the scientific literature of the citizens of nearby birds to provide accurate ideas.
There is plenty of room for Merlin’s Sound ID to grow. There are 10,000 birds, and the app only knows about 400 of them right now. The short sounding is difficult, as it sounds very similar between the animal species, while the program can err on some of the more frequent background sounds. But just as the demonstration is progressing, so too is the machine learning and the feasibility of the program.
Van Horn was delighted with the potential for mass exhibition and machine learning. He plans to use the species in other parts of the Cornell Lab of Ornithology, such as in bird sanctuaries with a fixed voice. Weber said he may have been able to use the metaphor to describe the birds’ flight over cities over a long period of time during their migration, and they may have used the genre to identify bird movies. Van Horn also told me that he was considering bias and other ethics in machine learning, pointing out that the switch was made for wildlife only, made using only what users agreed to give Cornell via eBird, and operates the user’s phone without sending data to Cornell.
The fact that there is a well-known knowledge in one of the most popular bird tools will be a welcome experience for many birds, and if I have tried, I can say with certainty that it works well. Experienced birdwatchers may still find that their ears are a little more accurate than the app, but for me, the tool was a great addition to my bird identification tool.
[ad_2]
Source link
