Diff-SSL-G-Comp: Towards a Large-Scale and Diverse Dataset for Virtual Analog Modeling

2025-04-06

AI Summary

Neural-network-based virtual analog (VA) models for dynamic range compression suffer from poor cross-parameter and cross-input generalization due to scarce and low-diversity training data. Method: We introduce the first large-scale, high-diversity real-hardware recording dataset specifically designed for SSL 500 G-Bus compressors—comprising 2,528 hours of audio from 175 unmixed tracks and 220 parameter combinations, spanning multiple genres, instruments, tempi, and keys. This is the first open-source real-recording dataset tailored for dynamic range compressors, accompanied by a unified benchmarking framework supporting black-box, gray-box, and white-box VA modeling, as well as ablation studies via curated subsets. Contribution/Results: Experiments demonstrate substantial improvements in modeling fidelity across diverse cross-parameter and cross-input evaluation scenarios for multiple open-source VA models. The dataset, codebase, and interactive demos are fully open-sourced.

Technology Category

• Machine Learning: Learning on the Edge & Model Compression
• Cognitive Modeling & Cognitive Systems: Affective Computing
• Computer Vision: Diffusion Models for Vision

Application Category

• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• User Modeling, Personalization and Recommendation: Practical large-scale studies of user experience
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Virtual Analog (VA) modeling aims to simulate the behavior of hardware circuits via algorithms to replicate their tone digitally. Dynamic Range Compressor (DRC) is an audio processing module that controls the dynamics of a track by reducing and amplifying the volumes of loud and quiet sounds, which is essential in music production. In recent years, neural-network-based VA modeling has shown great potential in producing high-fidelity models. However, due to the lack of data quantity and diversity, their generalization ability in different parameter settings and input sounds is still limited. To tackle this problem, we present Diff-SSL-G-Comp, the first large-scale and diverse dataset for modeling the SSL 500 G-Bus Compressor. Specifically, we manually collected 175 unmastered songs from the Cambridge Multitrack Library. We recorded the compressed audio in 220 parameter combinations, resulting in an extensive 2528-hour dataset with diverse genres, instruments, tempos, and keys. Moreover, to facilitate the use of our proposed dataset, we conducted benchmark experiments in various open-sourced black-box and grey-box models, as well as white-box plugins. We also conducted ablation studies in different data subsets to illustrate the effectiveness of improved data diversity and quantity. The dataset and demos are on our project page: http://www.yichenggu.com/DiffSSLGComp/.

Problem

• Lack of large-scale diverse dataset for Virtual Analog modeling
• Limited generalization in neural-network-based Dynamic Range Compressor modeling
• Need for benchmark experiments across different parameter settings

Innovation

• Large-scale diverse dataset for SSL 500 G-Bus Compressor
• Recorded 2528-hour audio with 220 parameter combinations
• Benchmarked black-box, grey-box, and white-box models