search

sp-uhh / sgmse

Score-based Generative Models (Diffusion Models) for Speech Enhancement and Dereverberation
MIT License
521 stars 76 forks source link

which checkpoint can be applied to bandwidth expansion tasks? #53

Closed huhuqwaszxedc closed 3 weeks ago

huhuqwaszxedc commented 3 months ago

Thank you for sharing. I am very interested in the performance of the proposed model in bandwidth extension tasks. Which checkpoint could be used?

julius-richter commented 3 weeks ago

None of the checkpoints are trained for bandwidth extension. Instead, they are trained on speech enhancement tasks. However, as we have discussed in [1], score-based generative models used for speech enhancement also perform well for bandwidth extension.

[1] Jean-Marie Lemercier, Julius Richter, Simon Welker, Timo Gerkmann, "Analysing Diffusion-based Generative Approaches versus Discriminative Approaches for Speech Restoration," IEEE Int. Conf. Acoust., Speech, Signal Process. (ICASSP), Rhodes, Greece, Jun. 2023.