New Season, New Tools: Yellowhammer Monitoring with BioacousticAI

Who is Yellowhammer?

The Yellowhammer (Emberiza citrinella) is a widespread songbird across Eurasia, with introduced populations in New Zealand. It is mostly non-migratory, with partial exceptions for some North – Eastern European populations.

It is known for its bright yellow plumage and striking, rhythmic song, often described as sounding like “a little bit of bread and no cheese.”

What makes this species particularly interesting for bioacoustic research is that the song contains both individually unique features and geographically meaningful components. The fast rhythmic ‘A little bit of bread and no’ part is the individually unique ‘first phrase’, where each male repeats a combination of types that act like an individual ID, as it remains stable over time. The latter long whistle-like ‘cheese-please’ part is the ‘dialect phrase’, shared between males within the same locality, which allows us to estimate the geographic patterns of the song distributions. For more details on their songs, you can visit YH World!.

About the projects

Since 2011, Yellowhammer has been one of the main study species for the Bioacoustic group of Charles University in Prague, where many projects have been conducted. Manual recordings of individual birds have been done since 2016, and the males and their territories have been identified using their vocal and visual cues.

We joined the project last year, expanding the horizons of the decade-long work, collecting data from a well-established study area in the suburbs of the Czech Republic. Now we are building a complete, long-term dataset where males are identified over multiple years based on their songs alone, to provide a profound basis for further work.

Along with continuing the legacy, this year we’re incorporating a major new element: passive acoustic monitoring (PAM) using autonomous recording devices. 35 Audiomoth from Open Acoustic Devices (OAD) have been deployed to cover a wide range of territories to continuously record Yellowhammer interactions throughout their breeding season.

Alongside, in collaboration with Dodotronic, we are testing a new recorder prototype with interactive applications and improvements in recording range and quality, and sound source localization with a multi-channel approach.These new practices will help provide greater spatiotemporal coverage without the need for constant human presence in the field. 

By combining manual directional recordings with this new passive monitoring network, we aim to compare the effectiveness of all methods and to develop and assess AI approaches to identify individuals from PAM data. Further on, we plan to expand our work to understand the population dynamics by observing interactions between territorial holders and floaters (non-breeding males), dive into the song culture intercourse between and within dialect borders to understand the distribution of phrases and dialects, and much more.

Can We Recognize Individual Animals by Their Voices?

Most bioacoustic studies today focus on identifying species from recordings. A particularly exciting and emerging field is Individual Acoustic Monitoring (IAM), which allows us to distinguish individuals of the same species based on their unique sounds – also known as acoustic signatures. What makes IAM particularly valuable is that it offers a non-invasive method of tracking animals. Unlike traditional methods of  individual capture and marking, which can stress animals and affect their behavior, IAM relies on sound recordings, minimizing the impact on the monitored animals.

IAM has huge potential for ecological research. It can be used to track the dynamics of animals across time and space, estimate population sizes and movement patterns. The possibilities of IAM are truly exciting, offering a powerful tool to enhance monitoring wildlife and gain deeper insights into animal behavior and conservation.

Why Artificial Intelligence?

Artificial Intelligence (AI) is crucial for processing and analyzing large volumes of acoustic data. Using AI to automate the identification of song types and individual vocal repertoires is an innovative approach in itself, providing crucial data for further applications.

However, there are still many challenges to overcome in applying AI to bioacoustic monitoring. For example, signal overlap with other sound sources can make it difficult to isolate and identify specific vocalizations. Additionally, factors such as Signal-to-Noise Ratio (SNR), the balance between the desired sound and background noise, pose further complications, especially in environments where noise levels are high.

The recently launched BioDcase challenge uses our Yellowhammer dataset to explore solutions and revisit the well-known problem of automatic detection of birdsong for Tiny Hardware (BioDCASE 2025).

We’re excited to contribute to this growing project and explore the potential of non-invasive, AI-powered bird monitoring. Stay tuned for more updates from the field!