STRUNK

Well .. making "models" for 1 NPC, splitting it in normal, soft and loud, in tortoise TTS takes me 3 days and produces 3 1.6 Gb models. It's just too much. Besides that, tortoise TTS with all the NPC voice models could be hosted for instance on discord, but one should have a dedicated high spec computer running 24/7 to make it available for everyone at any time. So that is kinda not realistic : P A friend of mine has made a Discord bot that runs LLM's from her own pc, and she can also run stable diffusion/sdxl and probably also TTS. I think she wouldn't mind sharing her discord bot/app code if someone is really intrested to use it for this purpose.

@TerrificTom

That was a nice and creepy story, and the atmosphere was very good too. The build had some flaws, but thanks for putting this together as a halloween suprise : )

Ok, I like to try that : )

Yes I still do, but .. I know they are AI generated ofc. but also I have edited audio and vocals so much in my life that I maybe listen to audio in a different way. Non the less, even this "home made" TTS that can run on your own pc/mac can be usefull in some cases, and will only get better. And if you can have the NPC's say really funny things in a mission, I guess no one will care that is was generated : P

For the sake of comparison, the same clip with background audio: NPC.mp3 It sounds robotic .. that keeps being true.

I picked up a conversation between some NPC's regarding our goabouts on this forum! NPCConversation.ogg

F5-TTS does take .ogg but it doesn't work with drag and drop, but does work when you click upload and select the .ogg file.

Yes the results can vary a lot but this F5 model is so fast I think it might still be a lot faster to generate the same thing a lot of times and finaly pick one that is good enough, then training voice models in tortoise-TTS and still have to generate multiple times to get a good result. After what I have been trying for almost a week now, TTS AI could be fun to use in some cases, preferably with background audio in the scene, to have NPC's react to specific situation, but not for real dialogue or cutscenes (yet).

Here is a demo page where you can test it yourself: https://huggingface.co/spaces/mrfakename/E2-F5-TTS I don't know if it directly takes .ogg but it does take .wav. (Batch) convert ogg(s) to wav(s) with vlc player works very well.

I have installed F5-TTS with pinokio: https://pinokio.computer/ It's remarkable how fast this model is and how easy to use. Just find some longer voice clip and use it as reference audio. Under Multi-Speech you can upload more voice clips and give them tags/Speech Type Names, of use this for different voices alltogether, then simply start the scentence with {your tag}. {your tag2} etc. I took 4 clips from the moor tagging them soft, normal, angry and shout, then making them say the same line: YourCamel.mp3 It works very fast and the quality is very much the same as I got out of tortoise-TTS, but seems not as good in keeping the character of the voice.

F5-TTS is a different, non autoregressive model. The moor voice was done with tortoise-TTS.

The cloning of the characteristics of the voice works quite nice. I selected sets of audio clips by how "loud" the speech is, Loud, Normal and Soft (speaking up, speaking normal, speaking soft). I didn't figure out how to get the best audio quality, without weird "audio artifacts". But as demonstrated down here, the moor certainly sounds like the moor: moorLoud_00000.wav

There is a new TTS : https://github.com/SWivid/F5-TTS Should be a lot better with "emotion".

I think it can run on ios and linux but that might not be as simple as the window install: https://github.com/neonbjb/tortoise-tts https://git.ecker.tech/lightmare/tortoise-tts