A Machine-Learning Approach to Application of Intelligent Artificial Reverberation

Abstract

We propose a design of an adaptive digital audio effect for artificial reverberation, controlled directly by desired reverberation characteristics, that allows it to learn from the user in a supervised way. The user provides monophonic examples of desired reverberation characteristics for individual tracks taken from the Open Multitrack Testbed. We use this data to train a set of models to automatically apply reverberation to similar tracks. We evaluate those models using classifier f1-scores, mean squared errors, and multi-stimulus listening tests.

A History of Audio Effects

“…Within this field, there has been a large focus on the use of some signal analysis approaches to automate, or directly control, some parameters of a preexisting audio effect. This can be performed by machine learning from data [153,154], performing some curve fitting or mapping to some higher-level parameters [155,156] or using some direct signal analysis to control an audio effect parameter directly [157,158].…”

Perceptual Evaluation and Analysis of Reverberation in Multitrack Music Production

“…Song and mix order was fully randomized, and all mixes were scaled to equal loudness according to [40]. At the end of the fragment, playback would loop to 3 For two songs permission to disclose artist and song name was not granted. the start of that fragment.…”

“…The ability to predict the desired amount of reverberation with a reasonable degree of accuracy has applications in automatic mixing and intelligent audio effects [2,3], novel music production interfaces (e.g., various mappings of low-level parameters to more perceptually relevant parameters or terms [4,5]), and compensation of listening conditions [6].…”

“…In contrast to other important mix engineering toolssuch as level [7,8], panning [9], EQ [10], and dynamic range compression [11,12]-to date only one attempt at automatic control of reverberators has been made [3]. Very little work is available on novel, more intuitive interfaces for reverb [13,14] and mapping terms to its parameters [4,5].…”

“…In some cases, the number of reverberator parameters were limited, often taking a restricted range or set of values [19][20][21], and applied to a single (type of) source sample [22][23][24]. In [3,25] the parameter values considered were set by unskilled participants using unfamiliar tools and inferior listening environments. Finally, the results of several parameter adjustment tests are not validated through perceptual evaluation [26][27][28].…”