SAGA-SR: Semantically and Acoustically Guided Audio Super-Resolution
Authors:
- 3 May 2026
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1706-1710
Abstract
Versatile audio super-resolution (SR) aims to predict high-frequency components from low-resolution audio across diverse domains such as speech, music, and sound effects. Existing diffusion-based SR methods often fail to produce semantically aligned outputs and struggle with consistent high-frequency reconstruction. In this paper, we propose SAGASR, a versatile audio SR model that combines semantic and acoustic guidance. Based on a DiT backbone trained with a flow matching objective, SAGA-SR is conditioned on text and spectral roll-off embeddings. Due to the effective guidance provided by its conditioning, SAGA-SR robustly upsamples audio from arbitrary input sampling rates between 4 kHz and 32 kHz to 44.1 kHz. Both objective and subjective evaluations show that SAGA-SR achieves state-of-the-art performance across all test cases. Sound examples and code for the proposed model are available online1.Keywords
All Related Versions
This publication has 15 references indexed in Scilit:
- Stable Audio Open2025
- FlashSR: One-step Versatile Audio Super-resolution via Diffusion Distillation2025
- Pixel-Aware Stable Diffusion for Realistic Image Super-Resolution and Personalized Stylization2024
- AudioLDM 2: Learning Holistic Audio Generation With Self-Supervised PretrainingIEEE/ACM Transactions on Audio, Speech, and Language Processing, 2024
- Audiosr: Versatile Audio Super-Resolution at Scale2024
- Scalable Diffusion Models with Transformers2023
- InstructPix2Pix: Learning to Follow Image Editing Instructions2023
- High-Resolution Image Synthesis with Latent Diffusion Models2022
- Towards Robust Speech Super-ResolutionIEEE/ACM Transactions on Audio, Speech, and Language Processing, 2021
- Look, Listen, and Learn More: Design Choices for Deep Audio Embeddings2019