yaaat 0.1.7

Interactive bioacoustic annotation tool for measuring vocalizations

YAAAT: Yet Another Audio Annotation Tool

Interactive bioacoustic annotation tool for measuring vocalizations.

Features:

1. Changepoint Annotator, for marking temporal onset, offset, and changepoints in vocalizations. Useful for describing rapid fluctuations and identifying nonlinear phenomena.
2. Peak Annotator, for marking dominant frequency peaks on the power spectrum. Useful for describing spectrally complex vocalizations.

Changepoint Annotator	Peak Annotator

Installation

Via Pip (probably easiest)

pip install yaaat

From Source

git clone https://github.com/laelume/yaaat.git
cd yaaat
pip install -e .

Usage

Launch the Application

yaaat

Opens a tabbed interface with both annotators. Includes test audio files to get started. For some reason, auto-load is a little buggy, so clicking Load Audio Files and selecting the included test_audio yourself lets the interface work as-intended.

Use in Python Scripts

from yaaat import ChangepointAnnotator, PeakAnnotator
import tkinter as tk

# Launch changepoint annotator
root = tk.Tk()
app = ChangepointAnnotator(root)
root.mainloop()

# Or launch peak annotator
root = tk.Tk()
app = PeakAnnotator(root)
root.mainloop()

Getting Started

1. Click Load Audio Directory to select files or Load Test Audio to explore test audio
2. Choose where to save annotations (existing, new, or default directory)
3. Click on the spectrogram to add annotation points
4. Click Finish Syllable when done with annotation
5. Move between files using Next/Previous buttons
6. Annotations auto-save on file navigation or Finish syllable

Navigation & Features

• Intuitive real-time interactive visualization with zoom, pan, and keycommand + mousewheel navigation

• Visualize harmonics with adjustable multipliers and draggable bounding boxes

• JSON annotations saved per-file to minimize corruption

• Mark and track unusable files

• Adjust spectrogram resolution for accuracy comparison

• TODO: implement ranking system for annotation quality; inject as learning feedback mechanism

Requirements

• Python ≥3.8 (built using 3.11)
• numpy
• matplotlib
• librosa
• scipy
• natsort
• sounddevice
• soundfile

License

MIT License - Copyright (c) 2025 laelume

Contributing

Contributions welcome! Please open an issue or submit a pull request. I'm especially interested in talking to people about using this in their existing AI workflows, so please feel free to reach out !!