The advent of deep learning has significantly influenced various fields, extending its impact across diverse domains. One notable application is its use in monitoring rare birds through their songs.

Differentiating between birds by their songs has become more accessible due to the availability of numerous mobile applications and software for ecologists and the general public. However, a major problem occurs when identification software comes across a species of bird with which it is unfamiliar or for which there are few reference recordings.

Ecologists and conservationists face the problem of monitoring some of the world’s rarest birds. To overcome this problem, researchers at the University of Moncton, Canada, have developed ECOGEN, which can generate lifelike bird sounds to enhance the samples of underrepresented species. These can then be used to train audio identification tools used in ecological monitoring.

Generating audio poses several challenges, including the substantial number of samples required for synthesis. Different formats are utilized for processing audio files, and many of these representations result in a loss of information, which complicates the production of high-quality audio samples. The waveform representation, which records sound pressure amplitude in the time domain, emerges as one of the most prevalent formats that maintains information integrity without loss.

To tackle this, ECOGEN has created novel instances of bird sounds to improve AI models. Essentially, ECOGEN enables the expansion of sound libraries for species with limited wild recordings without harming the animals or necessitating additional fieldwork.

The researchers found that adding synthetic bird song samples produced by ECOGEN to a bird song identifier improved bird song classification accuracy by 12% on average. One of the lead researchers, Dr. Nicolas Lecomte, underlined the urgent need for automated instruments, like acoustic monitoring, to track changes in biodiversity brought on by notable worldwide fluctuations in animal populations. However, thorough reference libraries are frequently absent from current AI models used for species identification in acoustic monitoring.

The researchers emphasized that creating synthetic bird songs can contribute to the conservation of endangered bird species and provide valuable insight into their vocalizations, behaviors, and habitat preferences.

Dr. Lecomte said that the tool could benefit other types of animals as well and said while ECOGEN was developed for birds, they are confident that it could be applied to mammals, fish, insects, and amphibians.

ECOGEN operates by transforming bird song recordings into spectrograms, visual representations of sounds. Subsequently, it generates new AI images based on these spectrograms, thereby augmenting the dataset specifically for rare species with limited recordings. These newly generated spectrograms are then converted back into audio format to train bird sound identification models. In this study, the researchers utilized a dataset comprising 23,784 wild bird recordings sourced globally, encompassing 264 different species.

Check out the Paper. All credit for this research goes to the researchers of this project. Also, don’t forget to join our 33k+ ML SubReddit, 41k+ Facebook Community, Discord Channel, and Email Newsletter, where we share the latest AI research news, cool AI projects, and more.

If you like our work, you will love our newsletter..
The post Meet ECOGEN: A Novel Deep Learning Approach Designed to Generate Realistic Bird Songs for Biologists and Ecologists appeared first on MarkTechPost.