Bird Audio Detection Challenge: Task specification
This page specifies how to submit your files. For general information see https://machine-listening.eecs.qmul.ac.uk/bird-audio-detection-challenge/
You will be given a set of 10-second audio files (44.1 kHz mono WAV files). For each file, your system needs to decide whether it contains any birds or not. This decision is expressed as 0 or 1. We encourage probabilistic, confidence-weighted or “fuzzy” decisions so your system is encouraged to output numerical values between 0 and 1 inclusive, to express the strength of decision.
Your system should output a results file in the following CSV format:
filename1,decision1 filename2,decision2 ...
Two columns, separated by a comma (and rows separated by a “newline”). The public “training” data contains an annotation file in this format. Note that the public “training” data has only 0 or 1 as the annotation, while we encourage your submission to make use of the continuous range between 0 and 1. There should be no additional whitespace in the file, except for the newlines.
You can use the public “training” data for training, or simply for checking the performance of your existing algorithm. (Not all detectors require training.) Please do not use data from outside the challenge. The system should be automatic; please do not manually modify the decisions.
You should produce decisions for the “testing” dataset. When you upload these, the system will tell you the score your system attains for a small portion of the testing dataset. This gives you some idea of how well the method might perform in the final test – but it is only an approximate guide! We have designed the challenge so that systems which generalise well (i.e. they avoid “overfitting”) should be the best performers in the final analysis.
You may submit multiple outputs – one per day, with an overall maximum of 20 per team – until the challenge deadline.
To be eligible for a prize, the method used needs to be described so that the work can potentially be adopted by others. The easiest way to do this is to publish your method as open-source software. However, if your approach is based on proprietary or sensitive methods, a good description of the method may suffice. The organisers reserve the right to decide which submissions meet these requirements.