Predicting binaural colouration using VGGish embeddings

Thomas McKenzie; Alec Wright; Dan Turner; Pedro Llado

Predicting binaural colouration using VGGish embeddings

Thomas McKenzie, Alec Wright, Dan Turner, Pedro Llado

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Abstract

An initial feasibility study is presented exploring the use of a pre-trained feature extractor, designed for large-scale audio classification, applied to the task of predicting the colouration between binaural signals. A multilayer perceptron (MLP) is trained to predict binaural colouration using feature embeddings obtained from the VGGish network and data from five previously conducted listening tests. The evaluation compares seven versions of the network, each trained using different data augmentation methods, to three existing signal processing methods: basic spectral difference (BSD), log. spectral distance (LSD) and an auditory model for predicting binaural colouration (PBC-2). Results show that while the MLP networks
are comparable to BSD and LSD, further work is needed to compete with the more accurate PBC-2; such as using specific audio features relevant for colouration

Original language	English
Title of host publication	AES International Conference on Artificial Intelligence and Machine Learning for Audio
Publisher	Audio Engineering Society
Pages	1-10
Number of pages	10
Publication status	Published - 2 Sept 2025

Access to Document

McKenzie2025AESPredictingAccepted author manuscript, 256 KBLicence: Creative Commons: Attribution (CC-BY)
McKenzie2025AESPredictingFinal published version, 255 KB

https://aes2.org/publications/elibrary-page/?id=23007

Cite this

@inproceedings{ba368ddde16f42da856cc264d5e19148,

title = "Predicting binaural colouration using VGGish embeddings",

abstract = "An initial feasibility study is presented exploring the use of a pre-trained feature extractor, designed for large-scale audio classification, applied to the task of predicting the colouration between binaural signals. A multilayer perceptron (MLP) is trained to predict binaural colouration using feature embeddings obtained from the VGGish network and data from five previously conducted listening tests. The evaluation compares seven versions of the network, each trained using different data augmentation methods, to three existing signal processing methods: basic spectral difference (BSD), log. spectral distance (LSD) and an auditory model for predicting binaural colouration (PBC-2). Results show that while the MLP networks are comparable to BSD and LSD, further work is needed to compete with the more accurate PBC-2; such as using specific audio features relevant for colouration",

author = "Thomas McKenzie and Alec Wright and Dan Turner and Pedro Llado",

year = "2025",

month = sep,

day = "2",

language = "English",

pages = "1--10",

booktitle = "AES International Conference on Artificial Intelligence and Machine Learning for Audio",

publisher = "Audio Engineering Society",

address = "United States",

}

TY - GEN

T1 - Predicting binaural colouration using VGGish embeddings

AU - McKenzie, Thomas

AU - Wright, Alec

AU - Turner, Dan

AU - Llado, Pedro

PY - 2025/9/2

Y1 - 2025/9/2

N2 - An initial feasibility study is presented exploring the use of a pre-trained feature extractor, designed for large-scale audio classification, applied to the task of predicting the colouration between binaural signals. A multilayer perceptron (MLP) is trained to predict binaural colouration using feature embeddings obtained from the VGGish network and data from five previously conducted listening tests. The evaluation compares seven versions of the network, each trained using different data augmentation methods, to three existing signal processing methods: basic spectral difference (BSD), log. spectral distance (LSD) and an auditory model for predicting binaural colouration (PBC-2). Results show that while the MLP networks are comparable to BSD and LSD, further work is needed to compete with the more accurate PBC-2; such as using specific audio features relevant for colouration

AB - An initial feasibility study is presented exploring the use of a pre-trained feature extractor, designed for large-scale audio classification, applied to the task of predicting the colouration between binaural signals. A multilayer perceptron (MLP) is trained to predict binaural colouration using feature embeddings obtained from the VGGish network and data from five previously conducted listening tests. The evaluation compares seven versions of the network, each trained using different data augmentation methods, to three existing signal processing methods: basic spectral difference (BSD), log. spectral distance (LSD) and an auditory model for predicting binaural colouration (PBC-2). Results show that while the MLP networks are comparable to BSD and LSD, further work is needed to compete with the more accurate PBC-2; such as using specific audio features relevant for colouration

UR - https://aes2.org/publications/elibrary-page/?id=23007

UR - https://aes2.org/events-calendar/2025-aes-international-conference-on-artificial-intelligence-and-machine-learning-for-audio/

UR - https://www.aes.org/openaccess/policy/

UR - https://www.aes.org/publications/proceedings/

M3 - Conference contribution

SP - 1

EP - 10

BT - AES International Conference on Artificial Intelligence and Machine Learning for Audio

PB - Audio Engineering Society

ER -

Predicting binaural colouration using VGGish embeddings

Abstract

Access to Document

Other files and links

Fingerprint

Cite this