Easybird: A Python Toolkit for Bird Activity Detection

Repo

realzza/easybird

Introduction

Easybird is a neural network-based python library for bird activity detection (BAD). It helps users detect whether a waveform contains bird sounds or not. The detection can be performed on a single waveform and multiple waveforms.

Installation

Install with pip

pip install easybird 

Manual Installation

First download the wheel file of easybird package.
Next, install the .whl file manually.

pip install easybird-0.0.2-py3-none-any.whl

Interface

easybird has a clean and succint interface to interact with the users.

From Single wav

from easybird import detection 

result = detection.from_wav('bird.wav', noise_thres=0.5)

Output

print(result)
# (True, 0.9996312260627747)

From Multiple wavs

from easybird import detection

results = detection.from_wavs(['bird1.wav','bird2.wav','bird3.wav'], noise_thres=0.5)

Output

print(results)
# [(bird1, True, 0.99963122), (bird2, True, 0.37834975), (bird3, True, 0.87340939)]

Highlights

• Detect bird activities from single wav
• Detect bird activities from multiple wavs
• Confidence score availble
• Automatic sample rate adaption

Data

The data used to train the bird activity detector is from DCASE 2018 Task 3. Since the label files for the Evaluation datasets are not released yet, we used the Development datasets only. The three datasets we utilized are listed below.

• freefield1010 [labels] [dataset]
• warblrb10k [labels] [dataset]
• BirdVox-DCASE-20k [labels] [dataset]