Description
GROSS DATA: 38m38s NET DATA: 37m33s EPOCHS: 540 (24,840 steps) METHOD: RVC v2 with mangio-crepe The prognosticator of prognosticators! Give him liberty or give him death! Here is your host, the man they call Ghost. as long as you ignore his crypto grifts, questionable content, fake rage, and maybe the occasional mantra of his granny being a pie ass woman. This is not only an extension of the previous version's dataset by adding more data I had sitting on my hard drive (was tinkering with what datasets work best), but a COMPLETE RETRAIN, utilizing mangio-crepe at 128 hops. Originally was gonna train to 600 epochs, but charts flatlined/showed no new decrease at around the 500 mark. The end result is (in my opinion and testing) a much cleaner and more coherent sounding output, with less voice cracks vs the first version! crepe is recommended for this model, set to 64-128 hops. Albeit this voice wasn't originally a singing voice, singing outputs turned out VERY well. Your mileage may/will vary, but you do you. SAMPLES < V1 <
Comments

GROSS DATA: 38m38s NET DATA: 37m33s EPOCHS: 540 (24,840 steps) METHOD: RVC v2 with mangio-crepe The prognosticator of prognosticators¹! Give him liberty or give him death! Here is your host, the man they call Ghost. || ¹as long as you ignore his crypto grifts, questionable content, fake rage, and maybe the occasional mantra of his granny being a pie ass woman. || This is not only an extension of the previous version's dataset by adding more data I had sitting on my hard drive (was tinkering with what datasets work best), but a COMPLETE RETRAIN, utilizing mangio-crepe at 128 hops. Originally was gonna train to 600 epochs, but charts flatlined/showed no new decrease at around the 500 mark. The end result is (in my opinion and testing) a much cleaner and more coherent sounding output, with less voice cracks vs the first version! crepe is recommended for this model, set to 64-128 hops. Albeit this voice wasn't originally a singing voice, singing outputs turned out VERY well. Your mileage may/will vary, but you do you. # SAMPLES https://discord.com/channels/1089076875999072296/1140104722301329548/1141343799612813352 <https://huggingface.co/tridenT98/GhostPolitics/resolve/main/GhostPoliticsV2.zip> V1: || <https://huggingface.co/tridenT98/GhostPolitics/resolve/main/GhostPolitics.zip> ||

Got a feeling the appreciation of this, as well as the context, will be lost on a lot of people unfortunately. ||Though it's totally understandable if you'd rather not hear this old hambone sing your favorite songs lol|| Good stuff.

Thank you very much.

how could I have gone without using this for a sample? here you go!

As this model has been updated/retrained, I will be changing the entire OP. This post contains the original OP's contents, and whatever information was given for the first version of mentioned model. Additionally, all samples above apply to V1 of this model, and updated samples will be given after this post, and any samples linked to the OP will be deleted, minus their mobile friendly forms (of which I can't toss). Any new samples here on out will be given the title of GhostV2. TITLE: Ghost (True Capitalist Radio/The Ghost Show) | RVC v2 | 600 Epochs || The prognosticator of prognosticators*, Here is your host, the man they call Ghost.|| ||Trained on 17 minutes (net) of data, taken from a very recent episode (on youtube).|| ||As the data doesn't have him sing, harvest is recommended, however crepe can give results on some tracks. Your mileage will vary on crepe and singing applications. ||

Ghost (True Capitalist Radio/The Ghost Show) | RVC v2 | 540 Epochs (UPDATED)

# SAMPLES (V2)
<@230131114671341568> Has anyone used this for Radio Graffiti?

Can't try on RG when copeler doesn't wanna do it (we all know he's gonna cave in and return to doing it like he always does)
:^)

SLASH OPEN THE mephitic dendum
Add a comment
Samples
Pitch
More to explore
Loading more