S3RL [ This Beat Has Got Me ]

# Last Update : <t:1707862174:R> > - **Model URL :** > - https://huggingface.co/rayzox57/S3RL_RVC/resolve/main/ThisBeatHasGotMeParts_v2_Ov2_750e.zip > - **Version :** > - :firev2: RVC V2.0 - Ov2 Super > - **Pitch Extraction Algorithm :** > - RMVPE > - **Epochs - Steps :** > - 750 - 3.750k > - **Dataset :** > - ~ 00:00:33 [ S3RL - This Beat Has Got Me - Acapella ] > - Enhanced Audio > - **Recommended Usage :** > - Cover > - **Search Feature Ratio :** > - 0.75 > - **Pitch :** > - ( + 4 = Man ) - Depend Base Voice > - You can adjust if you find better version # Previews : > - **Preview_Cover.wav :** > - *Provide* : Mustard - Ballin [ My DIY Acapella ] > - *Contains External Effects* : Yes > - *Pitch* : +4 > - *Feature Ratio* : 0.75

Reference : https://www.youtube.com/watch?v=-SnUylW_HY4

send raw vocals (without effects)

the raw preview ?

no it have reverb

raw mean without reverb eq

etc

wai

t

im idiot

yes

that's what I'm saying, you want the preview I put above without any effect (as if it came out of applio) right? :kekw: (im too slow)

what this one have reverb

I always mention in my description whether a preview has "effects" or not, but since I didn't make a preview on a TTS (where I don't use effects) I zapped it and sent a version without effects. So the one without effects is just above this message.

did u denoise i think i can hear noise in some places (or maybe its just voices in the head )

I don't understand anything, you're asking me without effects, this is the file as it comes out in applio, without going through audition.

what this is without effects

but i think i can hear noise in some places

no reverb, no denoise, nothing

i was talking about dataset

did u denoise dataset

no

rip i have to delete it if u want to reupload it denoise dataset and retrain it :C

I can assure you there's nothing more to denoise/eq. The noise you hear must come from the fact that the acapella I've set as input on applio isn't probably perfect. As for the dataset, it couldn't be cleaner since it comes from the acapella files supplied by S3RL (which, of course, I've sorted out to avoid a chorus of parts that could ruin the whole thing), provided we don't use them commercially.

but you said yourself that you didnt use denoise for dataset

Because there's no need for them, S3RL supplies the acapellas ready-made.

noise from this

Yes noise on the preview, and it shows that the dataset and the model are bad or contains noise ? No, it just shows that extracting acapella from music has never provided perfect results, and that if you put a voice model over it you'll get artifacts, but if you sing over it with a perfect mic then there won't be any unwanted noise.

just listen

Yes send in the RVC output straight from gradio without any audio processing

And no that is simply not true. While, it is true that if ur Acapella is of poor quality that u may get artifacting but that really doesn't matter because mostly it depends more on the model rather than the input audio it self. So when u haven't de noised ur models properly u get more noise

S3RL has already pre-processed the audios (denoise+eq ...), I took the audio and cut out all the unwanted parts.

well then, send me a sample of the dataset in dms

np