This site presents our research concerning the testing of algorithms of detection of clicks in audio recording.
Performance of Perceptible Impulsive Disturbance Detection Methods based on Auditory Models
paper submitted to JAES
The Matlab implementation of the algorithms can be downloaded here (for the reviewers).
Comparison of some click detection algorithms using listening tests resuls
F. Rund, V. Vencovský, and M. Semanský, “An Evaluation of Click Detection Algorithms Against the Results of Listening Tests,” J. Audio Eng. Soc., vol. 69, no. 7/8, pp. 586-593, (2021 July.). https://doi.org/10.17743/jaes.2021.0020
This paper evaluates the ability of several algorithms to detect impulse distortions (clicks) in audio signals. The systems are evaluated against data from a listening test conducted using real audio signals provided by a vinyl manufacturer. Some of the signals contained clicks due to damage during the manufacturing process. An evaluation of click detection algorithms against listening test results focuses on the ability of the click-detection algorithms to detect perceptible clicks. The results presented in this paper show that an algorithm that employs a hearing model detected audible clicks with a lower false detection rate than the other algorithms in the test and that the wavelet transform–based algorithm with a dynamic threshold outperformed the other algorithms.
The Matlab implementation of the algorithms including the used sound samples can be downloaded here. Please, cite the above referenced paper, if you use the data.
A. Ozdogru, “Comparison of Autoregressive Model Based Perceptible Click Detection Algorithms.” In: Proceedings of the International Student Scientific Conference Poster 26/2022. Praha: CTU. Faculty of Electrical Engineering, 2022. p. 1-4. vol. 1. ISBN 978-80-01-06992-9.
In this paper, Autoregressive (AR) model based click detection algorithms are compared in terms of their ability to detect only the perceptible impulsive noise (clicks) contained in the audio signals taken from damaged vinyl records. All the algorithms are using the same principle of detection but they are all improved differently in the mentioned literature. The test audio signals are previously classified to be containing or not containing clicks. Comparison criterion is based on a previously introduced custom function, defined through the correct and the false detection ratios. Run time comparisons of the algorithms are also made. Results showed that a Simple AR model algorithm performed the best, both in detection and in run time, amongst the compared AR algorithms.
M. Semanský, Systém pro detekci impuzních vad mechanického analogového záznamu využívající model slyšení (System for Impulse Noise Detection in Gramophone Records using Human Hearing Model) Master Thesis. 2019-06-12 Czech Technical University in Prague, Faculty of Electrical Engineering.