Recording Environment and Microphone Recognition in Cymbal Audios Using Machine Learning

Although machine learning techniques are commonly used in audio forensic analysis to identify recording devices and environments, the music industry can also benefit from this type of application, mainly because it is a little-explored approach. In this sense, this work aims to use this type of approach to assess whether different recording conditions - specifically microphones and environments - contribute to increasing the variability of audio signals. A dataset of cymbal audio recordings was analyzed using a logistic regression model combined with Recursive Feature Elimination with Cross-Validation (RFECV) to evaluate the classification of microphones and environments. Results indicate that the recording environments significantly impacted signal variability, achieving classification accuracy of 99%, whereas the influence of microphone types was minimal, with classification performance around 52%.