search

sp-uhh / sgmse

Score-based Generative Models (Diffusion Models) for Speech Enhancement and Dereverberation
MIT License
454 stars 69 forks source link

RTF seems too high, any idea to improve it? #12

Closed NormonisPing closed 2 months ago

NormonisPing commented 1 year ago

As above, training diffuse model with my own data, the quality is impressive but rtf is bad.

julius-richter commented 2 months ago

One way to reduce the RTF is proposed in our follow-up work: Single and Few-Step Diffusion for Generative Speech Enhancement