New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Comments
Probably... Got a squismellow like that I think.
Oh, I have heard German spoken in some movies. Never Duthc though. Hence, I was curious.
Always nice. I just had 2 sandwiches.
What's that? SOme kinda of a toy like Labubus?
He won't feel a thing.
Were rats involved in making these sandwiches?
Oh... Too bad.
Oh my god, that's so cruel, sara.
No, any good.?
Why is that though? Aint that a good thing?
Agreed.
It is a fun even, that's all I can say.
It takes dedication.... A lot of it. When I was younger I used to get up at 4.
Glad to have yes.
Thank God that I keep clearing my notifications from time to time. Else, I would end up with something like Ralf had.
Yah, it is easier to do it when you are younger. But, I guess, it's still possible. Some studies claim a habit can be built if you are consistent with it for 21 days.
You could be right. Kidsters messing about with make-up way too much these days. I blame SM.
Yah, me as well.
I like working actually.... Probably not a popular stance
Oh yes. SM has mastered the art of selling artificial beauty with their products, god knows sourced from where and by who and for what.
Hyping copium
I cna relate to that. I actually love what I do, I opted for it. Its just that a gaming PC is too much to give up on though
Sorry, we only do hypium here.
Glorp
Hey buddy, I am in the present. So, how is the internship going? I am typing up the comment for ralf on the TTS thingy right now.
Cool. Lunch break is almost over so I will get back to work soon. But I can't, since my cat is sleeping on my lap
I agree
@ralf Over the past few months, I researched a few TTS models and here are my findings that might help you.
These are the models I found best for most purposes:-
1. https://github.com/index-tts/index-tts - Imo, the best out there, but, no clear licence information. So, can't be sure if it can be used for commercial projects. One of the best features here is that it can also very reliably do emotions with weightage. There are several emotions that it supports by default. I tried mailing their team regarding clarification on the licence terms, but, they never replied. So, I moved onto the next one I found most suitable for my use case.
2. https://github.com/SWivid/F5-TTS - A very close second. Doesn't have emotion control and such. But, I guess, it can be trained to do that using multiple voices and using specific datasets using the custom toml file they specify on their Github. I did try that though. But, didn't find emotion control very smooth. Sometimes, teh voice is just shouting. But, cloning is prefect, I'd dare sya close to a 100% match.
Training your own voice
Both of the above models, esp. F5 has a training pipeline within the original source code that you can use very easily. For preparation of the dataset, i.e., to record your voice, I recommend https://github.com/rhasspy/piper-recording-studio. It has readmymade dataset for many popular accents. You can also add your own, which I do recommend to make sure there is good phoneme coverage. Remember, most of these models use pinyin characters, so phoneme coverage is important.
As for the dataset size, it depends, if you wanna train from scratch, you will need thousands of hours of raw voice data. But, if you can work by pre-training an already available model, I recommend F5_TTS instead of E2_TTS. Might take a bit of trial and error to find what workds for you best. If pre-training an already available model, you just need an hour of data, will work even with a few minutes of data, but, I do not recommend that.
Also, use a good condensor mic and maybe some denoising software for the best reults.
If you have any doubts, let me know. I will try to clarify according to my knowledge. Please understand I am no expert in this matter. EVerything I have written above is based on what I researched in the past few months and I even tested a few other models to find which one would be best for me like Orpheus, Dia, etc. Hope this helps someone.
Very different:)
Oh, you are WFH today? That's good. Anyways, posted my findings on TTS model.