The goal of the bioacoustics on tiny hardware task is to develop an automatic classifier of birdsong that complies with the resource constraints of low-cost and battery-powered autonomous recording units.

The evaluation set is now available:

Task 3 Evaluation

Description

The next generation of autonomous recording units contains programmable chips, thus offering the opportunity the opportunity to perform BioDCASE tasks. On-device processing has multiple advantages, such as high durability, low latency, and privacy preservation. However, such “tiny hardware” is limited in terms of memory and compute, which calls for the development of original methods in audio content analysis.

In this context, task participants will revisit the well-known problem of automatic detection of birdsong while adjusting their systems to meet the specifications of a commercially available microcontroller. The challenge focuses specifically on the detection of Yellowhammer bird vocalizations using the ESP32-S3-Korvo-2 development board.

The primary challenge is striking the optimal balance between classification accuracy and resource usage. While conventional deep learning approaches might achieve high accuracy, they may not be feasible on embedded hardware. Participants must explore techniques such efficient architecture design, and optimized feature extraction to create a solution that performs well within the hardware constraints.

A baseline implementation is provided as a starting point, which participants can modify and improve upon. Solutions will be evaluated based on classification performance, model size, inference time, and memory usage.

Note: We encourage participants to buy the ESP32-S3-Korvo-2 development board to test their solutions. The board is available for purchase on various online platforms (e.g, at DigiKey). However, the competition can be completed without the board, as the evaluation will be performed on a hidden test set using the baseline system by the organizers.

Dataset

The BioDCASE-Tiny 2025 competition uses a specialized dataset of Yellowhammer bird vocalizations. Key features include:

• 2+ hours of audio recordings
• Songs from multiple individuals recorded at various distances (6.5m to 200m)
• Recordings in different environments (forest and grassland)
• Includes negative samples (other bird species and background noise)
• Dataset split into training, validation, and evaluation sets

The training set contains recordings from 8 individuals, while the validation set contains recordings from 2 individuals. An additional 2 individuals are reserved for the final evaluation.

The dataset is organized as follows:

Development_Set/
├── Training_Set/
│   ├── Yellowhammer/
│   │   └── *.wav (Yellowhammer vocalizations)
│   └── Negatives/
│       └── *.wav (Other species and background noise)
└── Validation_Set/
    ├── Yellowhammer/
    │   └── *.wav (Yellowhammer vocalizations)
    └── Negatives/
        └── *.wav (Other species and background noise)

• Files under Yellowhammer/ contain Yellowhammer vocalization (target species for detection)
• Files under Negatives/ contain negatives samples which include vocalization from other species, as well as background noise recordings extracted for the full recordings.

The dataset is available for download on Zenodo:

BioDCASE-Tiny 2025 Dataset (4 GB)

The evaluation set is now available here

Evaluation and Baseline System

We provide a baseline system that includes a complete pipeline for audio processing, feature extraction, model training, and deployment to the ESP32-S3-Korvo-2 development board. The baseline system is designed to be easily adaptable for participants to build upon.

You can find the baseline system in the GitHub repository: BioDCASE-Tiny 2025 Baseline System

Submissions will be evaluated based on both classification performance and resource efficiency:

Classification Performance

• Average precision: The average value of precision across all recall levels from 0 to 1.

Resource Efficiency (measured on the ESP32-S3-Korvo-2 development board)

• Model Size: TFLite model file size (KB)
• Inference Time: Average time required for single audio classification, including feature extraction (ms)
• Peak Memory Usage: Maximum RAM usage during inference (KB)

Ranking

Participants will be ranked separately for each one of the evaluation criteria.

A baseline system is provided in the GitHub repository, including a complete pipeline for audio processing, feature extraction, model training, and deployment to the ESP32-S3-Korvo-2 development board.

External Data Resources

The competition focuses on the provided Yellowhammer dataset. External data use may be regulated according to the official competition rules

Rules and Submissions

1. Solutions must be deployable on the ESP32-S3-Korvo-2 development board
2. Models must process audio from the onboard microphone array in real-time
3. Submissions must include .h5 model files and pipeline_config.yaml in a zip file
4. Solutions will be evaluated on a hidden test set using the baseline system
5. Participants must cite the competition framework and dataset in any publications

A complete YAML metadata file with submission details should be provided. Here is an example Lastname_task3_1.meta.yaml:

# Submission information
submission:
  # Submission label
  # Label is used to index submissions, to avoid overlapping codes among submissions
  # use the following way to form your label:
  # [Last name of corresponding author]_[Abbreviation of institute of the corresponding author]_task[task number]_[index number of your submission (1-4)]
  label: Lastname_Institute_task3_Index

  # Submission name
  # This name will be used in the results tables when space permits
  name: Tiny Bioacoustics baseline

  # Submission name abbreviated
  # This abbreviated name will be used in the results table when space is tight, maximum 10 characters
  abbreviation: tinybioac

  # Submission authors in order, mark one of the authors as corresponding author.
  authors:
    # First author
    - lastname: Lastname
      firstname: Firstname
      email: email@email.edu                          # Contact email address
      corresponding: true                             # Mark true for one of the authors

      # Affiliation information for the author
      affiliation:
        abbreviation: CU
        institute: College University
        department: Computer Science
        location: Boring, Denmark

    # Second author
    - lastname: Lastname2
      firstname: Firstname2
      email: email@email.com                           # Contact email address

      # Affiliation information for the author
      affiliation:
        abbreviation: UC
        institute: University College
        department: Biology
        location: Boring, Maryland, USA

        #...


# System information
system:
  # The meta data provided here will be used to do meta analysis of 
  # the submitted system for the Bioacoustics for Tiny Hardware task.
  # Use general level tags, if possible use the tags provided in comments.
  # If information field is not applicable to the system, use "!!null".
  description:

    # Acoustic representation
    acoustic_features: spectrogram   # e.g one or multiple [MFCC, log-mel energies, spectrogram, CQT, PCEN, ...]

    # Data augmentation methods
    data_augmentation: !!null             # [time stretching, block mixing, pitch shifting, ...]

    # Embeddings
    # e.g. VGGish, OpenL3, ...
    embeddings: !!null

    # Machine learning
    # What kind of loss function was used during model training?
    training_objective: binary cross entropy         # e.g one or multiple [None, binary cross entropy, L2, L1]

    # Model compression
    # What kind of model compression techniques were used, if any?
    model_compression: None # e.g one or multiple [None, prunning, knowledge distillation, ...]

    # External data usage method
    # e.g. directly, embeddings, pre-trained model, ...
    external_data_usage: !!null

  # URL to the source code of the system [optional, highly recommended]
  source_code:   

  # List of external datasets used in the submission.
  external_datasets:
    # Dataset name
    - name: !!null
      # Dataset access url
      url: !!null
      # Total audio length in minutes
      total_audio_length: !!null            # minutes

# System results 
results:
  # Full results are not mandatory, but for through analysis of the challenge submissions recommended.
  # If you cannot provide all result details, also incomplete results can be reported.
    
    # Classification Performance
    average_precision: 0.5
    
    # Resource Efficiency on the ESP32-S3-Korvo-2 development board
    resource_efficiency:
      model_size: 10 # in (KB)
      inference_time: 100 # in (ms)
      peak_memory: 256 # in (KB)
    

Citation

If you use the BioDCASE-Tiny framework or dataset in your research, please cite the following:

Framework Citation

@misc{biodcase_tiny_2025,
  author = {Carmantini, Giovanni and Förstner, Friedrich and Isik, Can and Kahl, Stefan},
  title = {BioDCASE-Tiny 2025: A Framework for Bird Species Recognition on Resource-Constrained Hardware},
  year = {2025},
  institution = {Cornell University and Chemnitz University of Technology},
  type = {Software},
  publisher = {GitHub},
  journal = {GitHub Repository},
  howpublished = {\url{https://github.com/birdnet-team/BioDCASE-Tiny-2025}}
}

Dataset Citation

@dataset{yellowhammer_dataset_2025,
  author = {Morandi, Ilaria and Linhart, Pavel and Kwak, Minkyung and Petrusková, Tereza},
  title = {BioDCASE 2025 Task 3: Bioacoustics for Tiny Hardware Development Set},
  year = {2025},
  publisher = {Zenodo},
  doi = {10.5281/zenodo.15228365},
  url = {https://doi.org/10.5281/zenodo.15228365}
}

Support

If you have questions please use the BioDCASE Google Groups community forum, or contact the task organizers at: stefan.kahl@cornell.edu