-
|
Brilliant work @Purfview with XXL! The added features are greatly useful. I was so excited when I saw you have added Kim-2 and Silero-3. Very good choices. I just tested XXL beta running on RTX3600 with very good results. A quick question: in the past I have used Kim-2 in UVR5. In UVR5 the setting prameters for Kim-2 (MDX) are: Segment Size, and Overlap. In XXL beta the (available) setting parameter is mdx_chunk. I was wondering (a) does mdx_chunk somehow correspond to the segments_size? (b) if available, would it be an idea to expose the overlap parameter in CLI? |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 1 reply
-
|
I meant new post(aka comment) instead of reply, not new thread. :)
"Segment Size" is related to mdx_chunk, here I think it corresponds to seconds, there I dunno what it means, as I remember there Segment Size default is "256" and it corresponded to ~5-8(?) seconds. |
Beta Was this translation helpful? Give feedback.
-
|
Btw, Kim_vocal_2 is pretty good but current the best vocal extraction is BS RoFormer model, I'll check if that can be implemented. |
Beta Was this translation helpful? Give feedback.
I meant new post(aka comment) instead of reply, not new thread. :)
"Segment Size" is related to mdx_chunk, here I think it corresponds to seconds, there I dunno what it means, as I remember there Segment Size default is "256" and it corresponded to ~5-8(?) seconds.
Here is no such thing as "Overlap", it's not direct implementation from UVR.