Is there a significant difference between 40K and 48K during training?

[hexed gull]

I heard that within UVR, everything over 40K will be cut out, and that 40K is better. Is this true?

And after refining the song in DAW, I extracted it to 48K and saved it. Is there any problem if I learn 40K in that state? Should I extract it according to 40K?

[covert schooner]

I heard that within UVR, everything over 40K will be cut out, and that 40K is better. Is this true?

Depends on the UVR model.

And for the second question, it's fine if you use lower

Honestly I always train on 32k to prevent noise, and 48k aint worth it

[hexed gull]

Does this mean that it is okay to start learning with a 40K algorithm after extracting it to 48K?

[covert schooner]

yeah

[hexed gull]

Does learning at 32K eliminate a lot of noise? Doesn't it ruin the tone a lot?

[covert schooner]

doesn't ruin the tone, eliminates some of that noise

just plain better for most models

i'd say all models tbh

[hexed gull]

I will have to try making 40K and 32K based on the model I already made with 48K. Thank you.

[covert schooner]

Alrighty! Let me know how it goes :)

[hexed gull]

I tried learning at 32k, and the noise was definitely reduced, and the sound of the AI ​​vocals felt much thicker. Thank you for the good tip.

[covert schooner]

You're welcome!

[tender gull]

main question is the difference between 40K and 48K during training? Actually none. No one on Earth can hear the difference between 40K and 48K. But rvc wise, using proper pretrain from your dataset is a must, to lessen artifacting.

[lusty river]

most of the time it will end up being 32k, 40k is rare

you need to check with a spectrogram, Spek or this for example https://huggingface.co/spaces/TheStinger/Ilaria_Audio_Analyzer

look for the peak of the frequencies and double them

if they all reach 16k, select 32k