Automatic Audio Equalization with Semantic Embeddings

2026-07-26

AI Summary

This work addresses the problem of blind audio equalization—automatically restoring spectral characteristics without reference signals—by proposing a lightweight, semantics-guided deep learning approach. The method leverages semantic embeddings extracted from a pretrained model as backbone features and fine-tunes only a lightweight head network to predict log-mel spectrograms, from which inverse filters are derived for automatic equalization. The proposed framework achieves an effective balance between training efficiency and generalization capability, demonstrating strong performance on both music and speech signals while exhibiting robustness to noise and reverberation. Objective evaluations and subjective listening tests indicate that its performance closely approaches that of an ideal system using ground-truth log-mel spectrograms, highlighting its promising potential for practical applications.

Abstract

This paper presents a data-driven approach to automatic blind equalization of audio by predicting log-mel spectral features and deriving an inverse filter. The method uses a deep neural network, where a pre-trained model provides semantic embeddings as a backbone, and only a lightweight head is trained. This design is intended to enhance training efficiency and generalization. Trained on both music and speech, the model is robust to noise and reverberation. Objective evaluations confirm its effectiveness, and subjective tests show performance comparable to that of an oracle that uses true log-mel spectral features, indicating that the model accurately estimates the desired characteristics, with remaining limitations attributed to the filtering stage. Overall, the results highlight the potential of the method for real-world audio enhancement applications.

Problem

• audio equalization
• blind equalization
• log-mel spectral features
• semantic embeddings
• audio enhancement

Innovation

• semantic embeddings
• blind equalization
• log-mel spectral features
• lightweight neural head
• audio enhancement