Need help with where to put finetuned XTTS model in my directory, and what config settings to update

[fandango328]

I finetuned an xtts model today and renamed it to model.pth and placed it in my alltalk_tts\models\xtts\finetuned folder but I'm not sure that's right. There's a lot of documentation about how to conduct the finetuning training, but if it's mentioned what you're supposed to do with it once you have it trained.

I've made slight updates to the config.json and the confignew.json in my main alltalk_tts folder to try and remap to the new model. I've spent the last 2 days trying to troubleshoot this, and even did a full uninstall and reinstall to start fresh and I'm still struggling.

I'm also making things more difficult than they need to be, because I'm trying to remote access Alltalk from my Raspberry Pi4 using the api, and I just can't figure it out.

So far I have updated these two files:
confignew.json

tgwui

• enable deepspeed
• updated temp, repetition penalty, and generation speed values
• updated character voice to the new voice in .wav format

api_def

• updated api_legacy_ip_address to my local machines LAN IP address

config.json

I'm not sure what this started off as, but this is what GPT-4o told me to do with this file

{
  "host": "0.0.0.0",
  "port": 7851,
  "alltalk_host": "http://localhost:7851",
  "alltalk_path": "I:\\alltalk_tts",
  "max_queue_size": 10,
  "request_timeout": 30,
  "default_voice": "Laura.wav",
  "default_model": "xtts",
  "save_transcripts": true,
  "language": "en",
  "speed": 1.0,
  "temperature": 0.6,
  "models": {
    "xtts": {
      "model_path": "models/xtts/finetuned/model.pth", 
      "config_path": "models/xtts/finetuned/config.json",
      "vocab_path": "models/xtts/finetuned/vocab.json"
    }
  },
  "optimization": {
    "use_deepspeed": true,
    "use_low_vram": false,
    "generation_speed": 0.96
  }
}

If anyone can tell me what the correct directory location or setting updates I need to make that would be amazing. thank you!

[RenNagasaki]

Hey,
first of all if you only want to use alltalk via API, why integrate it into TGWUI?
I'd recommend installing it as standalone.

Did you use the alltalkbeta(v2)-branch or the main(v1)-branch?
I'd recommend using v2 since v1 is quite old and AFAIK not actively worked on anymore.

I don't recommend touching either of the configfiles (except to change the api and gradio ports) unless you know for sure what you're doing.

But going back to your inital question, all you have to do is: put the *.pth files into a folder below /models/xtts/
For example mine looks like this:
I:\Projekte\alltalk\models\xtts\xtts2.0.3-trained
And inside I have these files:

Then once you start alltalk go to either:

• API default URL: http://0.0.0.0:7851
• Gradio default URL: http://0.0.0.0:7852
  The ports are default, but if you changed them inside the confignew.json then you gotta change the upper accordingly.

Switch the dropdown inside available models. (The entrys are created like that: {parent folder (xtts)} - {sub folder (xtts2.0.3-trained)} )P.S. Simply use the xtts version, don't bother using apitts( it doesn't change anything AFAIK)


Don't forget to hit 'Change Model'/'Load Different Model'

I guess you want to use custom voices so don't forget to put samples into the voices folder since you've trained your own model.

Kind regards,
Ren