Hi! I'm looking to request a girl speaking model that can help me pass as a trans girl. Most if not all models that I've tried struggle with laughing, which is really off putting. I would like a model that makes me feel like I can be myself, without having to change how I am for the voice to sound fine. Many thanks
#Female Voice Model
amber turtle
edgy stump
Hey,
I’d be happy to give it a try, but just so you know—RVC usually has a hard time with laughter and other emotional sounds. I can’t promise it will sound completely natural, but I’m willing to see what we can get out of it.
amber turtle
Hey,
I’d be happy to give it a try, but just so you know—RVC usually has a hard ...
I did notice some models handle it better than others. Is this because of laughs being in the dataset, or is it like a luck factor that just feels better with the voice itself?
iron geyser
I did notice some models handle it better than others. Is this because of laughs...
this is a good question, i don't think laughing in the dataset improves them
afaik rvc can't do all type of laughs because the f0 estimator can't track the pitch in some scenarios
that but also because rvc is not a true voice cloning, is it spectogram cloning
the model learns spectograms, then converts them to .wav
so it doesn't know how to do emotions
amber turtle
Voice models that are often from a louder person tend to do them better than those soft voices
I don’t think loud is the correct way to describe them, but I’m sure you know what I mean
iron geyser
Back to what I was saying earlier, it’s possible that rvc can produce certain types of laughter in a "natural" way, but it’s not possible to achieve natural results with all types of laughter since rvc cannot learn emotions, it only learns pitch and spectrograms
some TTS can do emotions, like eleven, because they're actual voice cloning
so those models do learn different emotions during training, rvc sadly does not
amber turtle
Mhm it’s unfortunate that it can’t
And I’m guessing there won’t ever be an improvement to rvc that could breakthrough this?
Or would a completely new thing have to be introduced!
iron geyser
i think it can be improved if a better f0 than rmvpe is created
amber turtle
The soft voices just don’t work for me because it stops you from being able to be yourself
iron geyser
oh yea avoid asmr models, they're trained mostly on noise data (whispering, a lot of breaths, etc etc) rvc doesn't like that
something better than rvc have to be made in order to improve laughs yea nothing can be done in the current rvc arch to improve them, besides maybe a new f0 like i said before
amber turtle
I’ve looked into making my own but I don’t really have a strong computer to do training. It can barely hang on to rvc
iron geyser
hmm you can try cloud training, lighting and kaggle are the most commonly used options here, colab too
amber turtle
something better than rvc have to be made in order to improve laughs yea nothing...
I’ll gladly wait for that day to come, but it does feel like this server isnt how it was like 6+ months ago
iron geyser
I’ll gladly wait for that day to come, but it does feel like this server isnt ho...
100% i agree with this
upgrading rvc is a quite hard task, while it's possible to make some things better, like more natural speech, it always have the big downside of asking for stronger gpus
rvc is also quite old now... it was made late 2022/early 2023
amber turtle
Back when it was new people were very amazed but it feels like normal people have caught on also
It isn’t improving at a steady rate like other ai things are
Which is very sad
iron geyser
i hope we can get something better soon, but for now rvc is all we have for casual usage