Description
Before Start - During the training process, I saw that a model already existed on weights.gg under the name "Unknown Singer (Singer of Like the Wind / The Most Mysterious Song On)", so I decided to go ahead anyway, as I really wanted to create my own version. - Despite efforts to separate the acapella from the instrumental, the style of the music made it difficult to perform 100% cleanly. But the result is cool! - The model only works with acapellas where the voice is strong. As an example of what I've been able to do to test the model: the Ballin' acapella gives a great result, whereas the Your Reality, ost acapella from DDLC gives a very poor result, as the voice is very timid and soft. Last Update : <t:1704482580:R> - Model URL Version RVC V2.0 - Pitch Extraction Algorithm RMVPE - Epochs - Steps 750 - 3.75k - Dataset ~ 00:00:33 [ Only the "main voice", or rather the most imposing one!] - Recommended Usage Cover, with a loud voice ! - Don't try soft covers, they don't seem to work at all. - Search Feature Ratio 0.75 - Pitch Logic Pitch ( -6 / 0 = Man ) - You can adjust if you find better version Previews Preview_Cover.wav *Contains External Effects* : Yes - *Pitch* 6 - *Feature Ratio* : 0.75 - *Additional Infos* : The chorus effect you can hear in the preview is native to the model, it's not an additional effect I've implemented ! The reverb only accentuates this effect
Comments
Before Start - During the training process, I saw that a model already existed on weights.gg under the name "Unknown Singer (Singer of Like the Wind / The Most Mysterious Song On)", so I decided to go ahead anyway, as I really wanted to create my own version. - Despite efforts to separate the acapella from the instrumental, the style of the music made it difficult to perform 100% cleanly. But the result is cool! - The model only works with acapellas where the voice is strong. As an example of what I've been able to do to test the model: the Ballin' acapella gives a great result, whereas the Your Reality, ost acapella from DDLC gives a very poor result, as the voice is very timid and soft. Last Update : <t:1704482580:R> - Model URL Version RVC V2.0 - Pitch Extraction Algorithm RMVPE - Epochs - Steps 750 - 3.75k - Dataset ~ 00:00:33 [ Only the "main voice", or rather the most imposing one!] - Recommended Usage Cover, with a loud voice ! - Don't try soft covers, they don't seem to work at all. - Search Feature Ratio 0.75 - Pitch Logic Pitch ( -6 / 0 = Man ) - You can adjust if you find better version Previews Preview_Cover.wav *Contains External Effects* : Yes - *Pitch* 6 - *Feature Ratio* : 0.75 - *Additional Infos* : The chorus effect you can hear in the preview is native to the model, it's not an additional effect I've implemented ! The reverb only accentuates this effect 🙂
Reference :
Add a comment
Samples
1. Singing
Male
English
2. Singing
Female
English
3. Singing (Dry)
Female
English
4. Singing (High)
Female
English
5. Singing 2
Male
English
6. Singing (Dry)
Male
English
7. Singing (Dry, High)
Male
English
Pitch
Weekly Metrics
More to explore
Loading more