ISCA Archive Interspeech 2025
ISCA Archive Interspeech 2025

A Neural Codec Approach for Noise-Robust Bandwidth Expansion

Xi Liu, Mu Yang, Szu-Jui Chen, John H.L. Hansen

Neural audio codec technology has recently attracted significant attention in various speech processing tasks due to its efficient quantized latent features. In this work, we introduce a novel approach that leverages a pre-trained neural codec network to perform both speech denoising and bandwidth expansion simultaneously. Specifically, we design a conformer-based deep neural network to predict clean codebook indices, which are then used by the pre-trained audio codec model to generate enhanced and bandwidth-expanded audio. We investigated several strategies for generating the clean indices and compared our approach with state-of-the-art methods on the Valentini-Botinhao noisy test set. Experimental results demonstrate that our method achieves performance comparable to leading approaches in noise-robust bandwidth expansion tasks while offering promising improvements in the quality and intelligibility of narrow-band signals. Audio samples are available.