Bird watching has become an increasingly popular hobby over the past few decades. Being able to identify birds by sight is important for any birder, but being able to identify birds by their calls and songs takes birding to the next level. iNaturalist is a popular app and website that allows users to identify plants, animals, insects, and more using computer vision technology and crowdsourcing from experts. But can iNaturalist accurately identify bird calls and songs?
What is iNaturalist?
iNaturalist is a joint initiative by the California Academy of Sciences and the National Geographic Society that connects users to nature and science. It allows users to record and upload photos or audio clips of plants, animals, fungi, and other organisms they encounter in the wild. These observations are then identified with the help of artificial intelligence and the iNaturalist community.
iNaturalist uses computer vision technology to suggest identifications for uploaded photos. It compares visual patterns, shapes, colors, and other features to images in its database to provide a species name. For audio recordings, it similarly analyzes audio waveforms and patterns to suggest potential matches. However, computer vision technology still has limitations, so final species IDs are provided by the iNaturalist community. Experts weigh in to confirm or correct computer suggestions, with research-grade observations receiving consensus from multiple users.
Bird Call Identification Challenges
Identifying birds by sight is already challenging enough. Factors like lighting, angle, obstructed views, age, color morphs, and individual variations can make it difficult to pin down an ID. Identifying birds by sound adds another layer of complexity. Here are some of the main challenges of identifying bird calls and songs:
Regional Dialects
Bird calls can vary significantly between regions, similar to regional dialects in human language. A robin in California may sound different than a robin in New York. Background species and habitat sounds also differ. This makes developing a single universal sound profile for each species difficult.
Individual Variation
There is variation between individuals of the same species, based on factors like age, sex, and learning. Some birds may be better at mimicking sounds or manipulating their vocalizations.
Mimicry
Some birds excel at mimicry and incorporate sounds from other birds or even man-made objects into their songs. This can confuse both human and computer listeners.
Complexity
Bird vocalizations can be highly complex, with multiple notes, trills, harmonics, and frequency modulations. Teasing apart the signature sounds to identify species is tricky.
Context Dependence
Birds can have different sounds for different contexts like mating, territorial defense, flock communication, alarm calls, etc. A single bird may have a repertoire of a half dozen or more distinct sounds.
Rarity
Rare or endangered species may not have as many sound recordings available for algorithms to learn from compared to common backyard birds.
Background Noise
Environmental conditions like wind, rain, flowing water, and calls from other animals can mask key identifying features of bird sounds.
Strengths of iNaturalist Bird Call Identification
While identifying bird sounds is difficult, iNaturalist does have some strengths that may allow it to provide reasonably accurate IDs in many cases:
Large Audio Database
iNaturalist has a database of hundreds of thousands of bird call recordings collected from users around the world. This big data helps the algorithm find matches.
Machine Learning
The algorithm “learns” over time by analyzing more recordings and getting feedback from users on correct IDs. This allows it to continuously improve.
Verification from Experts
Even if the computer makes an incorrect ID suggestion initially, expert users can correct it to provide the right species name.
Community Consensus
Research-grade observations require agreement from multiple users, rather than just a single ID. This consensus improves accuracy.
Regional Filters
The app allows filtering by location to match dialect. Users can also browse the Sound Library by region to hear local versions of bird sounds.
Multiple Recordings
Users can upload multiple recordings of the same bird from different angles or contexts. This provides more sound samples to analyze.
Sonic Details
Advanced audio analysis can extract identifying features like frequency, pitch, cadence, tone quality, and more to aid identification.
Case Studies on iNaturalist’s Accuracy
How accurate is iNaturalist specifically for identifying bird calls in real-world use? Here are the results from two case studies testing its performance:
Case Study 1
A 2021 study published in the journal Ecological Solutions and Evidence tested iNaturalist’s bird call identification accuracy using audio recordings of 30 species from the southeastern United States. The samples included common backyard species like cardinals, woodpeckers, and chickadees.
They found that initial computer vision suggestions matched human expert identifications just 57% of the time. However, after community input, the final consensus identifications rose to 83% accuracy. Rare species and ones with high individual variation like Northern Mockingbirds performed worse.
Case Study 2
A 2022 study published in BMC Ecology and Evolution evaluated iNaturalist specifically for nocturnal flight calls of migrating birds. These are short, simple chip or buzz sounds birds make during migration. The researchers compiled 587 flight call recordings from 96 species.
Here, iNaturalist’s algorithm alone correctly identified the calls 94% of the time to species or species group. After community review, accuracy increased to 98%. They found identifiers were most accurate for calls that were loud, low-frequency, and included multiple notes.
Tips for Improving Accuracy
Based on user experiences and scientific evaluations, here are some tips to get the most accurate IDs when using iNaturalist for bird calls:
Use Good Recording Equipment
Make high-quality recordings with minimal background noise. Directional microphones can isolate the bird’s sounds.
Get Multiple Angles
Record calls from different sides of the bird if possible to capture subtleties.
Capture Diagnostic Features
For longer songs, include repeated phrases to aid identification.
Note Behavior and Context
Share notes on what the bird was doing (singing, calling, flying, etc.) to provide clues.
Verify the ID
See if multiple users agree on the species for research-grade consensus.
Compare Similar Species
Listen to recordings of easily confused birds to learn the differences.
Quiz Yourself
Test your ears by guessing species before looking at IDs.
Conclusion
Overall, iNaturalist shows promising accuracy at identifying bird calls, especially with multiple user verifications. While the computer vision algorithm alone may only get the species right 50-60% of the time, consensus from the community of experts boosts accuracy up over 80-90%. Challenging factors like individual variation, mimicry, and rare species still lower accuracy compared to visual IDs.
As the app gains more recordings and the machine learning algorithm improves, performance should continue to increase. User tips like getting quality recordings and noting bird behavior also help. While not perfect, iNaturalist is currently the most powerful tool available for crowdsourced bird call identification.