Exiled_Vizir

Yes that's a good idea, mainly if you have a recent gfx card, but if you have to use only your CPU, one thing to know is that Bark is way faster in CPU only mode than Tortoise (I cannot test for know but I tend to think that if if you have a Geforce it should also be faster in GPU mode), it has the cool thing of speaking language with accent, the down side is that the voices number is more limited and self create voice is much more difficult than in tortoise. And also the sample you generate cannot be very longwith Bark, between 10 and 20 seconds normally.

Hi, thanks for this detailed return, you nailed it, Sabrina has recent Italian origins so I try to make her speak english with some italian accent so I try to select voices sample with italian accent speaking english or italian. For Tortoise-tts I don't know if it's my samples that should be better or if the software itself "saxonize" voices a lot but the young Monica Bellucci and Sabrina Salerno english sample are fluent english but with strong italian accent and for Salerno Ita she speak italian directly but still they sound more american. The Bark-Ita voices are made with with another software (suno-AI/bark) and this one is in its own league because if you select an Italian voice, enter an english test it translate the text and says it in Italian with the Italian voice selected... But! If put an Italian sentence at the begining of the text then some english text after you get the english text not translated but with an Italian accent... let's say that I don't fully understand what I'm doing but normally:

Hi, glad to see that you still manage to make progress on your project, for the point 2, nanite skin should be customizable at run-time? For the point 1 and 3 I agree with Sexvision these are nice to have but it can be done in a second time if it's too time consuming, especially for the h5m format because if I understand well, it would useful only for H5 objects which have their OBJ file missing, no?

Since I didn't have ideas for something new lately, I decided to make a new installation of TK17, because my main install is bloated by too much add-ons, mods, etc. So I got a new, very light install but with some cool eyes candy (Hook5, Xbody) so to to challenge it I'm going to remake my first serious video I made long time ago on the garden: Forbidden Dream. Also I wanted to re-use what I learned with Grand Theft Arcanum (sound, voices) but Forbidden Dream is longer, maybe more dialog density... So making voices with Tortoise-TTS in CPU mode would be long... very long. Luckily a good fellow named Chapoly1305 or super Chiba make a fork of Tortoise-TTS compatible with Radeon card, with I can generate voice a lot faster, so I can try more things: I started an artificial voices casting for Sabrina the main character of Forbidden Dream, for now I have 6 voices for her. You can hear them here. If you are interested by that remake tell me which suit the better for her in the poll associated with this blog page. If you don't know this character you can see the old video here. The new look of Sabrina (and the less ghetto toilets):

For those who are interested to try tortoise-tts on a radeon or an Intel GPU, install pytorch on directML use this version of the project by Chapoly1305 (many thanks to him for the adaptation of the code on pythorch DirectML). Also if your cards has less than 16 Go of VRAM you will to add a --batch_size argument to reduce the size of the portion of the model loaded in VRAM. By default it seems that --batch_size=14 is used, when I used the CPU this configuration used ~16 Go of RAM on my system, to run the program on my gfx card with 10 Go of VRAM I used --batch_size=8, with that between 9 and 10 Go are used in the 2nd phase of the generation of sample. In the issues of the original projects some Geforce 1060 with 6 Go of VRAM report to use --batch_size=4 parameter. So the formula for the batch_size seems to be memory target - 2.

Hi, I continue my journey in the fabulous world of artificial voicing recently, the last time I spoke of my test of a torch-directml/rocm-hip-sdk5.6.1 version under linux or WSL which was approx. 7 time faster than CUDA on CPU under windows version. And I wasn't if that acceleration involve my GPU or not. So now I have the answer, the GPU was not used, but recently I find a new version and: And as you can see my non Nvidia GPU is used by tortoise-tts, but apart using more electricity, is it useful? Yes it seems because I generate a sentence of 7 words I used in my Grand Theft Arcanum video in 3 minutes versus 15 minutes torch-directml/rocm-hip-sdk5.6.1 under WSL. So for now we have Radeon 6700 which have approximatively 80/90% of the compute power of a geforce 1080 based on the information given by deluga503 in this thread.

Hi, you can create your own voiceset with your recorded voices then import it into the game (it's like an addon) with that tool. Once you import it, create or modify a model and change the voice to your new voiceset (it's in the last tab in the customizer). After that you can use it in the sequencer with your model. Or if you know a video editor you can use some video of TK17 you have recorded and use your voice sample in the video editor to add your speech in the TK17 video.

Hi hfg2, have you made some progress on the pose editor of your project or with the graphic pipeline?

Hi, is it poses from v7.5 or vX native poses? If it's v7.5 poses do you have activate the options to convert them automatically in v10 poses? If not you can try it and if they had been converted you can do the oposite: try tonot convert them to see if it changes something.