Reverberation features identification from music recordings using the discrete wavelet transform

This paper presents a method of extracting reverberation features from music recordings. First, we perform a short time Fourier transform to transform the audio signal into a 2D time-frequency representation in which reverberation features appear as blurring of spectral features in the time dimension. Employing image analysis method we may quantitatively estimate the amount of reverberation by transforming the STFT “image” to a wavelet domain where we can perform efficient edge detection and characterization. Experiments demonstrate that quantitative estimates of reverberation time extracted in this way are strongly correlated with physical measurements.

1. INTRODUCTION

When a sound is emitted in an enclosed space an audience will perceive the combination of the direct sound and the subsequent multiple reflections and absorption of the sound by the walls as reverberation. Humans possess astonishing capabilities for perceiving and appreciating a surrounding aural space [1], [2] and acoustic reverberation is a significant part of our everyday aural experience. In applications such as speech recognition or automated music transcription engineers regard reverberation as a nuisance and a great deal of the research literature is focused on how to remove or partly mitigate its effects [3]. On the other hand, for musicians centuries of tradition have led to an appreciation of musical space as an indispensible creative tool for creating music and many musical compositions are composed for specific aural spaces [4].