Best model for early reverb?

[10Elem]

Good day. Thanks to the developers for your work! I have a question. What is the best algorithm for removing early reverberation reflections at the moment? I still have hope of recording vocals at home. )
If not, are there any searches and developments in this direction?

P.S. MB Roformer - DeReverb-DeEcho 1 - maybe this is it, but I couldn't run it in UVR5, it gives an error. I installed MB Roformer - DeReverb-DeEcho 2, but it doesn't capture early reflections very well.

In general, if someone ever manages to create a model for removing early reverberation reflections, it will change the game. In this case, many vocalists will finally be able to record vocals at home and the need for a studio and a booth for recording vocals will disappear.

[deton24]

Mel-Roformer de-reverb by anvuew (a.k.a. v2/19.1729 SDR) | DL | config | Colab
is probably the best de-reverb for now and RX11's dialogue isolate for de-echo.

If you have at least 4GB VRAM, you can try decreasing chunk_size to 112455 to avoid errors related to low VRAM using the latest UVR beta Roformer version:
https://docs.google.com/document/d/17fjNvJzj8ZGSer7c7OFe_CNfUKbAxEh_OBv94ZdRG5c/edit?tab=t.0#heading=h.6y2plb943p9v
or use these models on MVSEP.com or x-minus.pro (faster)

[10Elem]

Mel-Roformer de-reverb by anvuew (a.k.a. v2/19.1729 SDR) | DL | config | Colab is probably the best de-reverb for now and RX11's dialogue isolate for de-echo.

If you have at least 4GB VRAM, you can try decreasing chunk_size to 112455 to avoid errors related to low VRAM using the latest UVR beta Roformer version: https://docs.google.com/document/d/17fjNvJzj8ZGSer7c7OFe_CNfUKbAxEh_OBv94ZdRG5c/edit?tab=t.0#heading=h.6y2plb943p9v or use these models on MVSEP.com or x-minus.pro (faster)

Thank you for your reply. The video memory is fine, my 8 gigabytes from 3070 Ti are enough. I use Mel-Roformer de-reverb by anvuew as the main one. So far, in my opinion, the best is really Mel-Roformer de-reverb by anvuew and, unlike any other algorithms, it already captures early reflections a little, which I have not observed in any previous de-reverberation algorithms.

But unfortunately, there is no algorithm yet that would really make the voice dry, completely removing the earliest reflections. I hope that the clear and brilliant minds of the developers will be able to reach this height. )

[deton24]

Probably, you should get more dry results with RX11. śr., 15 sty 2025 o 14:54 10Elem ***@***.***> napisał(a):

…

Mel-Roformer de-reverb by anvuew (a.k.a. v2/19.1729 SDR) | DL <https://huggingface.co/anvuew/dereverb_mel_band_roformer/resolve/main/dereverb_mel_band_roformer_anvuew_sdr_19.1729.ckpt> | config <https://huggingface.co/anvuew/dereverb_mel_band_roformer/resolve/main/dereverb_mel_band_roformer_anvuew.yaml> | Colab <https://colab.research.google.com/github/jarredou/Music-Source-Separation-Training-Colab-Inference/blob/main/Music_Source_Separation_Training_(Colab_Inference).ipynb> is probably the best de-reverb for now and RX11's dialogue isolate for de-echo. If you have at least 4GB VRAM, you can try decreasing chunk_size to 112455 to avoid errors related to low VRAM using the latest UVR beta Roformer version: https://docs.google.com/document/d/17fjNvJzj8ZGSer7c7OFe_CNfUKbAxEh_OBv94ZdRG5c/edit?tab=t.0#heading=h.6y2plb943p9v or use these models on MVSEP.com or x-minus.pro (faster) Thank you for your reply. The video memory is fine, my 8 gigabytes from 3070 Ti are enough. I use Mel-Roformer de-reverb by anvuew as the main one. So far, in my opinion, the best is really Mel-Roformer de-reverb by anvuew and, unlike any other algorithms, it already captures early reflections a little, which I have not observed in any previous de-reverberation algorithms. But unfortunately, there is no algorithm yet that would really make the voice dry, completely removing the earliest reflections. I hope that the clear and brilliant minds of the developers will be able to reach this height. ) — Reply to this email directly, view it on GitHub <#1700 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AIJ3EHGXXCV63U7GJ6QLUEL2KZSA5AVCNFSM6AAAAABVABUZMWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDKOJSHEYTQMBWGY> . You are receiving this because you commented.Message ID: ***@***.***>

[deton24]

Maybe you could also try out Dango. It's sounds very similar to RX11 Dialogue Isolate set to 5:
https://tuanziai.com/en-US/de-reverb (free 30 seconds)