Welcome to the Audio Fingerprinting Competition! Audio fingerprinting is crucial in various applications, such as music identification, content recognition, and copyright enforcement. In this competition, participants will have the opportunity to showcase their expertise in developing innovative algorithms for audio fingerprinting, contributing to the advancement of audio technology. The goal is to develop accurate and efficient algorithms for audio fingerprinting. Participants’ systems must accurately match a query audio clip to the corresponding audio track in a database. This task simulates real-world scenarios where audio queries are possibly distorted by noise and reverberation, and the content needs to be quickly and accurately identified.
Task: Participants must create a system capable of identifying the top five matching audio tracks from a reference database. These matches should correspond to a provided monophonic audio query.
Background: Audio fingerprinting systems mainly consist of two components:
The competition database consists of diverse audio tracks spanning various genres. The database is divided into two main parts:
Development database: Participants will have access to a database containing 8000 audio tracks (*.wav). These tracks are sampled at a rate of 48kHz and each audio clip spans 30 seconds. Alongside the audio tracks, users will also be provided with a collection of noise clips and Room Impulse Responses (RIRs) to facilitate the creation of their representation learning models. Users may split this database into subsets for training and validation purposes. Furthermore, the same database can be employed to assess the performance of retrieval techniques.
Test database: It consists of reference audio tracks (relatively larger) and query audio clips. The reference audio tracks form the basis for generating fingerprints, while the query audio clips are used to evaluate the matching performance of the participants’ systems. The query clips are 2 seconds long and intentionally distorted by adding noise and reverberation at different SNR(dB) levels.
.csv format
with first column representing the query_filename
and column 2 through column 6 representing song_id
of the top-5 similar matches in the reference database.We will fill this section with questions we receive and our answers.
For additional information, you can reach us via email at wissap2023@gmail.com