SignalTrain: Profiling Audio Compressors with Deep Neural Networks

Published in

arXiv.org(2019)

External Links:

Generate Graph

TL;DR

A data-driven approach for predicting the behavior of a given non-linear audio signal processing effect (henceforth "audio effect") using a deep auto-encoder model that is conditioned on both time-domain samples and the control parameters of the target audio effect.

Abstract

In this work we present a data-driven approach for predicting the behavior of (i.e., profiling) a given non-linear audio signal processing effect (henceforth "audio effect"). Our objective is to learn a mapping function that maps the unprocessed audio to the processed by the audio effect to be profiled, using time-domain samples. To that aim, we employ a deep auto-encoder model that is conditioned on both time-domain samples and the control parameters of the target audio effect. As a test-case study, we focus on the offline profiling of two dynamic range compression audio effects, one software-based and the other analog. Compressors were chosen because they are a widely used and important set of effects and because their parameterized nonlinear time-dependent nature makes them a challenging problem for a system aiming to profile "general" audio effects. Results from our experimental procedure show that the primary functional and auditory characteristics of the compressors can be captured, however there is still sufficient audible noise to merit further investigation before such methods are applied to real-world audio processing workflows.

Authors

Scott H. Hawley

2 papers

Benjamin Colburn

1 papers

S. I. Mimilakis

1 papers

References60 items

Pending

A Feature Learning Siamese Model for Intelligent Control of the Dynamic Range Compressor

Formant estimation and tracking: A deep learning approach.

Deep Learning for Tube Amplifier Emulation

End-to-end music source separation: is it possible in the waveform domain?

SignalTrain: Profiling Audio Compressors with Deep Neural Networks

Published in

arXiv.org(2019)

External Links:

Generate Graph

TL;DR

Abstract

Authors

Scott H. Hawley

2 papers

Benjamin Colburn

1 papers

S. I. Mimilakis

1 papers

References60 items

Pending

A Feature Learning Siamese Model for Intelligent Control of the Dynamic Range Compressor

Formant estimation and tracking: A deep learning approach.

Deep Learning for Tube Amplifier Emulation

End-to-end music source separation: is it possible in the waveform domain?

Modeling of nonlinear audio effects with end-to-end deep neural networks

Log Hyperbolic Cosine Loss Improves Variational Auto-Encoder

Dereverberation binaural source separation using deep learning

Time-Frequency Networks for Audio Super-Resolution

A disciplined approach to neural network hyper-parameters: Part 1 - learning rate, batch size, momentum, and weight decay

Efficient Neural Audio Synthesis

Adversarial Audio Synthesis

Visualizing the Loss Landscape of Neural Nets

Automatic differentiation in PyTorch

Virtual Analog Modeling of a UREI 1176LN Dynamic Range Control System

A Wavenet for Speech Denoising

Virtual Analog Modeling of Dynamic Range Compression Systems

Neural Audio Synthesis of Musical Notes with WaveNet Autoencoders

Tacotron: Towards End-to-End Speech Synthesis

Char2Wav: End-to-End Speech Synthesis

Remixing music using source separation algorithms to improve the musical experience of cochlear implant users.

Convolutional recurrent neural networks for music classification

WaveNet: A Generative Model for Raw Audio

Deep Neural Networks for Dynamic Range Compression in Mastering Applications

More Than 50 Years of Artificial Reverberation

Deep Residual Learning for Image Recognition

Fast and Accurate Deep Network Learning by Exponential Linear Units (ELUs)

Learning Spectral Mapping for Speech Dereverberation and Denoising

U-Net: Convolutional Networks for Biomedical Image Segmentation

Detection and Classification of Acoustic Scenes and Events

Joint Optimization of Masks and Deep Recurrent Neural Networks for Monaural Source Separation

Adam: A Method for Stochastic Optimization

End-to-end learning for music audio

Improved musical onset detection with Convolutional Neural Networks

Model-Based Inversion of Dynamic Range Compression

Digital Dynamic Range Compressor Design—A Tutorial and Analysis

Recent Advances in Real-Time Musical Effects, Synthesis, and Virtual Analog Models

ImageNet: A large-scale hierarchical image database

Numerical Methods for Simulation of Guitar Distortion Circuits

Parameter Estimation of Dynamic Range Compressors: Models, Procedures and Test Signals

Attack and Release Time Constants in RMS-Based Feedback Compressors

A Simple Weight Decay Can Improve Generalization

Digital Dynamic Range Compressor for Audio

An Improved Method of Audio Level Control for Broadcasting and Recording

Hack Audio (Audio Engineering Society Presents): An Introduction to Computer Programming and Digital Signal Processing in MATLAB (Routledge

Acoustic Scene Classification by Ensembling Gradient Boosting Machine and Convolutional Neural Networks

SGDR: Stochastic Gradient Descent with Restarts

“ANALYSIS AND SIMULATION OF AN ANALOG GUITAR COMPRESSOR,”

Physical audio signal processing : for virtual musical instruments and audio effects

Supplementary Materials, including audio clips and an interactive demo

“Kemper Amps,”

The system could be modified to predict one sample at a time, however our experience with this model has found this practice to be neither necessary nor helpful

the requirements of diﬀerentiability imposed by training via gradient descent optimization

Given the requirements of differentiability imposed by training via gradient descent optimization, one might expect the discontinuous nature of a switch to pose a problem for training

Source code and datasets accompanying this paper will be publicly released on GitHub.com pending acceptance for publication

Callbacks.one cycle -fastai

Initially we chose control settings according to a symmetric beta distribution,

We will make the dataset publicly available pending review

Field of Study

Computer ScienceEngineering

Journal Information

Name

ArXiv

Volume

abs/1905.11928

Venue Information

Name

arXiv.org

Type

URL

https://arxiv.org

Alternate Names

ArXiv

TL;DR

Abstract

Authors

References60 items

Pending

A Feature Learning Siamese Model for Intelligent Control of the Dynamic Range Compressor

Formant estimation and tracking: A deep learning approach.

Deep Learning for Tube Amplifier Emulation

End-to-end music source separation: is it possible in the waveform domain?

TL;DR

Abstract

Authors

References60 items

Pending

A Feature Learning Siamese Model for Intelligent Control of the Dynamic Range Compressor

Formant estimation and tracking: A deep learning approach.

Deep Learning for Tube Amplifier Emulation

End-to-end music source separation: is it possible in the waveform domain?

Modeling of nonlinear audio effects with end-to-end deep neural networks

Log Hyperbolic Cosine Loss Improves Variational Auto-Encoder

Dereverberation binaural source separation using deep learning

Time-Frequency Networks for Audio Super-Resolution

A disciplined approach to neural network hyper-parameters: Part 1 - learning rate, batch size, momentum, and weight decay

Efficient Neural Audio Synthesis

Adversarial Audio Synthesis

Visualizing the Loss Landscape of Neural Nets

Automatic differentiation in PyTorch

Virtual Analog Modeling of a UREI 1176LN Dynamic Range Control System

A Wavenet for Speech Denoising

Virtual Analog Modeling of Dynamic Range Compression Systems

Neural Audio Synthesis of Musical Notes with WaveNet Autoencoders

Tacotron: Towards End-to-End Speech Synthesis

Char2Wav: End-to-End Speech Synthesis

Remixing music using source separation algorithms to improve the musical experience of cochlear implant users.

Convolutional recurrent neural networks for music classification

WaveNet: A Generative Model for Raw Audio

Deep Neural Networks for Dynamic Range Compression in Mastering Applications

More Than 50 Years of Artificial Reverberation

Deep Residual Learning for Image Recognition

Fast and Accurate Deep Network Learning by Exponential Linear Units (ELUs)

Learning Spectral Mapping for Speech Dereverberation and Denoising

U-Net: Convolutional Networks for Biomedical Image Segmentation

Detection and Classification of Acoustic Scenes and Events

Joint Optimization of Masks and Deep Recurrent Neural Networks for Monaural Source Separation

Adam: A Method for Stochastic Optimization

End-to-end learning for music audio

Improved musical onset detection with Convolutional Neural Networks

Model-Based Inversion of Dynamic Range Compression

Digital Dynamic Range Compressor Design—A Tutorial and Analysis

Recent Advances in Real-Time Musical Effects, Synthesis, and Virtual Analog Models

ImageNet: A large-scale hierarchical image database

Numerical Methods for Simulation of Guitar Distortion Circuits

Parameter Estimation of Dynamic Range Compressors: Models, Procedures and Test Signals

Attack and Release Time Constants in RMS-Based Feedback Compressors

A Simple Weight Decay Can Improve Generalization

Digital Dynamic Range Compressor for Audio

An Improved Method of Audio Level Control for Broadcasting and Recording

Audio

Hack Audio (Audio Engineering Society Presents): An Introduction to Computer Programming and Digital Signal Processing in MATLAB (Routledge

Acoustic Scene Classification by Ensembling Gradient Boosting Machine and Convolutional Neural Networks

SGDR: Stochastic Gradient Descent with Restarts

“ANALYSIS AND SIMULATION OF AN ANALOG GUITAR COMPRESSOR,”

Physical audio signal processing : for virtual musical instruments and audio effects

Supplementary Materials, including audio clips and an interactive demo

“Kemper Amps,”

The system could be modified to predict one sample at a time, however our experience with this model has found this practice to be neither necessary nor helpful

the requirements of diﬀerentiability imposed by training via gradient descent optimization

Given the requirements of differentiability imposed by training via gradient descent optimization, one might expect the discontinuous nature of a switch to pose a problem for training

Axe-Fx II XL+

Source code and datasets accompanying this paper will be publicly released on GitHub.com pending acceptance for publication

Callbacks.one cycle -fastai

Initially we chose control settings according to a symmetric beta distribution,

We will make the dataset publicly available pending review

Field of Study

Journal Information

Name

Volume

Venue Information

Name

Type