>inb4 self promoting fag Im getting ready to package things up for anons and doing final touch ups. Before I do, I want to do a final check on things you might want?Requested in /lmg and vcg>GPTsovitsV2pro [done]>GPTsovitsV2pro training [done]>browser extension for TTS [done]>translator [not done, can't verify quality as a non Japanese speaker]Someone here said his GPTsovitsV2pro voices were among the highest quality he's had for realtime TTS, and I agree. I was so busy focusing on VRAM heavy voices like Fish2, Zonos2 Qwen3 TTS that I overlooked great realtime voices for bots as well as providing a simple integrated training tab for those engines. If anyone has a link to a perfect voice along with its training dataset, that would be very handy to go through and integrate into the guide. I found some perfect Japanese voices at https://booth.pm/ja/items?tags%5B%5D=GPT-SoVITS , but no training data.For the anon that tested the diarizer, Im working on using .vtt files from crunchycunts to get anime voice splitting from 70-90% accuracy to about 93-98% per episode. Once its done, point at an anime season, select experimental diarizer, give it a .vtt file using https://pastebin.com/URCqdnJa to grab, and it will annotate the .vtt file first to note every character it can and make refined cuts from that file instead of whisperX.I hit a scraping issue with crunchyroll called P-Kat-3002 that once I fix ill just scrape all of the .vtt files over a few months
Here's the current list of engines. Focus atm is TTS, will finish RVC and Okada 2.0 live stuff once Ive tested all the engines on long form tasks like podcasts and book conversion to hunt for bugs. Browser extension renderings pop up in history, but Ill add a toggle to turn that on/off
>>109609509you link doesnt work lmao retarded self-promo
>>109609563Yes, its empty dick head. Its not ready, that's literally what I said in the OP.Eat shit lol
will this allow me to do real time voice
>>109609643Yes. It has realtime streaming and can gen faster than you can hear with a few engines.If you try an engine and like it, it's not a bad idea to find that lighter engine on its own and make a separate lighter harness for it that doesnt use this frontend/backend at all though. Worst case, this is great for trying out a model before you build that model into something
>>109609509>things you might wantMight be out of your hands, but the more stably it still runs when straining the available VRAM/RAM, the better.
>>109610001Yeah Ive ran into that issue swapping between engines and not checking whats unloaded/loaded, so thats good, but not perfect.
>>109609509What is the difference between this and audio.cpp?
>>109610857A focus on character saving and auto binding I guess. Audio.cpp looks amazing
>>109609509>Someone here said his GPTsovitsV2pro voices were among the highest quality he's had for realtime TTSYeah that's me lol. Did you look at moonshinev2 for diarization? For a good dataset I recommend genshin impact voices, it's among the highest quality ootb: https://github.com/AI-Hobbyist/Genshin_Datasets
>>109609509Comfyui fucked up all audio implementations. I can't take this thread seriously if you use that shit
>>109611236>Comfyui fucked up all audio implementationshow?
>>109611244Noticably more artifact and shit quality compared to ggml impls. Also python can go eat a dick. Comfyorg is a saas company so they can eat a dick
>>109609509You should probably change the name since comfyorg is full of grifters
