About whitenoise performance

[sqummy]

Thanks for your awesome work! I have installed deepfilternet through pip and test some samples. The overall noise reduction effect is great, but in whitenoise scene, there are more residues in the speech, which leads to a poor subjective feelings. Have you noticed this phenomenon? I will attach the samples below. samples.zip

[Rikorose]

Hi, thanks for your interest in DeepFilterNet! The sample you provided has a very bad SNR (I guess <-10dB). Even though DeepFitlerNet2's performance is quite good, these bad SNRs are too challenging I guess. Do you have some other approaches that sound well given this SNR?

[sqummy]

The SNR of the previous zip file is indeed low, I replaced a zip file with a higher SNR. My main question is why the residue exists in speech, and in non-speech segments the effect is relatively better. Intuitively, whitenoise should be easy to deal with. samples_higer_snr.zip

[Rikorose]

I get a different output than you: femaleSNR20dB_DeepFilterNet2.zip This sounds much better.

Could you send me some logs to see if something went wrong when you processed the samples?

[sqummy]

[sqummy]

Is there something wrong?

[Rikorose]

Seems like DeepFilterNet 1 is loaded. I will push a new release that includes the new DeepFilterNet2 checkpoints soon. The new model should provide a better performance.

[sqummy]

Thanks

[Rikorose]

Hi I pushed a new release. Could you update with: pip install -U deepfilternet and try again?

[github-actions[bot]]

This issue is stale because it has been open 90 days with no activity. Remove stale label or comment or this will be closed in 7 days.