A high quality 3-minute and 37-seconds dataset.
Batch size: 4
Hop Length: 8
Original Pretrain
Sampled at 40k
21 messages · Page 1 of 1 (latest)
A high quality 3-minute and 37-seconds dataset.
Batch size: 4
Hop Length: 8
Original Pretrain
Sampled at 40k
To rate this model, either use /rate or click the buttons below.
I know, but in this case, Mangio-crepe outperformed RMVPE—simple as that.
yea but it dosent do any sense train with mangio crepe and inference with rmvpe
did you inference with mangio crepe?
btw @azure patio can you send me the opening exercise audio on dms plz?
I’ve made over a thousand models and learned the strengths and weaknesses of these algorithms through trial and error. I can usually tell which algorithm will work best with a dataset. RMVPE works best about 95% of the time, but Mangio-crepe can outperform it in specific cases, like with short, simple, high-quality datasets that haven’t undergone much processing. When I listened to this dataset, I knew Mangio-crepe had a high chance of sounding better. I also made an RMVPE version just to be sure, but it sounded worse—glitchy and tearing—while Mangio-crepe’s lower accuracy made the result sound smoother and more natural. Sorry for the long text, I tend to overcomplicate my thoughts.
I understand, Mangio crepe can be helpful if rmvpe adds like ''artifacts'' or ruins your model. Btw i'll use rmvpe on mangio crepe models for inference for more precision
becuase mangio crepe on inference isnt very precise and i think you used rmvpe for the inference
Oh yes, I used RMVPE for the samples. Now I understand what you meant.
Good! I'm making a model with applio's crepe (mangio crepe) with the same hop length u used for this model.
Let's see if it comes out well.
ov2
e
@azure patio im a little bit confused on this part
This model has been synced with Weights and is ready to use for free!