Tommyinnit(Singing Voice)(RVC V2)(RMVPE)(Original)(1100 Epochs - 7700 Steps)(32k)

[lunar iron]

Retrain of my Harvest Model of Tommyinnit
Used
MVSEP:(BS-Roformer - MelBand Karaoke - Reverb Removal BS)
RX 10(De Crackle - De Ess - Voice De Noise - Normalize -3.00 - EQ)
Audacity(Truncated Silence - Noise Gate - T De Esser)
Dataset:8 Minutes 30 Seconds
Trained & Tested On Applio(Kaggle)
Recorded Vocals From The Karaoke Stream(Twitch)
Mic Used From Tommyinnit:AKG P120 Studio Condenser XLR
https://huggingface.co/notiscBalbalEko/myrvcmodel/resolve/main/TommyinnitSingingRetrain_1100e_7700s.zip?download=true

[small mango]

Rate the model

To rate this model, either use /rate or click the buttons below.

[lunar iron]

Forgot to put this :p

[kind fern]

you tried klm 4.1? it may help the model when it reaches high notes :p

it's just a suggestion, I'm not forcing ya to use it

[lunar iron]

tommy's voice is raspy cuz of his mic that's why it voice cracks

[kind fern]

probably, rarely can be the pitch extraction algorithm (rmvpe) with the possiblity of voice cracks

voice cracks mostly appear on mangio-crepe / crepe but yea it can depend, or its ur dataset or its the pitch extraction :p

anyways 4 star model

[marble orchid]

is not because the pitch extraction, is because his model voice range is low and he is trying to make the model sing notes the dataset doesnt have

[kind fern]

thats why samples of dis model are with voice cracks?

[marble orchid]

yes

[kind fern]

he probably should use klm 4.1 to fix this if im not wrong

since klm 4 = more vocal range ?

[marble orchid]

more voice range true but models have voice cracks when the model tries to do a sound the dataset lacks, so klm is not gonna fix the voice cracks, is just going to make them appear less when the model is singing high notes (this is literally the intended purpose of klm)

[sage birch]

Model Ready ✅

This model has been synced with Weights and is ready to use for free!