EBEN: Extreme Bandwidth Extension Network

Accepted to ICASSP 2023
Submitted to IEEE TASLP

Julien Hauret, Thomas Joubaud , Véronique Zimpfer, Éric Bavu

Paper Notebook Code Presentation Prototype


We present Extreme Bandwidth Extension Network (EBEN), a Generative Adversarial network (GAN) that enhances audio measured with body-conduction microphones. This type of capture equipment suppresses ambient noise at the expense of speech bandwidth, thereby requiring signal enhancement techniques to recover the wideband speech signal. EBEN leverages a multiband decomposition of the raw captured speech to decrease the data time-domain dimensions, and give better control over the full-band signal. This multiband representation is fed to a U-Net-like model, which adopts a combination of feature and adversarial losses to recover an enhanced audio signal. We also benefit from this original representation in the proposed discriminator architecture. Our approach can achieve state-of-the-art results with a lightweight generator and real-time compatible operation.


Motivation of our work


(quiet conditions)


(quiet conditions)


(85dB noise)


(85dB noise)

Listen to in-ear, reference and enhanced audios





HifiGAN v3
Streaming Seanet

Inspect corresponding spectrograms by yourself !

Funny experiments

In this example, we trained a bandwidth extension model on a given language and inferred it on degraded audio in another language. In the obtained enhanced audio, you'll notice a strong "accent", based on the language the model was trained for. Speech quality metrics do not penalize this behavior and give similar scores to models trained in the same language as inference.


(by a model trained with english content)




(by a model trained with french content)



Perceptual Evaluation

In order to evaluate the performance of the trained models, we performed two MUSHRA listening tests with 88 (resp. 82) participants on sound quality and ease of understanding. Each participant had to listen to 7 different audios enhanced with 5 different approaches (+reference +low anchor + original corrupted audio). For each experiment, the MUSHRA procedure consisted in giving a note between 0 and 100. The results from this study are found below.

Speech quality metrics on test dataset

Frugality indicators

Approach Generator Discriminators
Audio U-net 71.0M 0
Hifi-GAN v3 1.5M 70.7M
Seanet 8.3M 56.6M
Streaming-Seanet 0.7M 56.6M
EBEN 1.9M 27.8M

Approach Inference mode latency for a single one-second sample on a NVIDIA GeForce RTX 2080 Ti
Audio U-net 37.5ms
Hifi-GAN v3 3.1ms
Seanet 13.1ms
Streaming-Seanet 7.5ms
EBEN 4.3ms

Approach Inference mode maximum memory allocation for a single one-second sample on a NVIDIA GeForce RTX 2080 Ti
Audio U-net 1117.3 MB
Hifi-GAN v3 22.2 MB
Seanet 89.2 MB
Streaming-Seanet 10.9 MB
EBEN 20.0 MB

EBEN hyperparameters

Name Learning rate Betas
Adam 3e-4 (0.5,0.9)

Batch size Temporal length Sampling rate
16 2 sec 16 kHz

Bands number (=decimation) Kernel size Beta
4 32 9

Reconstructive Adversarial
100 1

Generator Discriminators
LeakyReLU(negative_slope=0.01) LeakyReLU(negative_slope=0.2)


@article{hauret2023eben, title={EBEN: Extreme Bandwidth Extension Network applied to speech signals captured with noise-resilient microphones}, author={Hauret, Julien and Joubaud, Thomas and Zimpfer, Véronique and Bavu, Eric}, journal={Submitted to ICASSP 2023} doi = {10.48550/ARXIV.2210.14090}, year = {2022}, url = {https://arxiv.org/abs/2210.14090}, copyright = {Creative Commons Attribution 4.0 International} }