No matter what operating system you're on you're already playing with AI here. Ask Claude to help you get it working. give him the readme instructions and show him the startup script, tell him what OS you're on. It works. You just have to make it work for your OS and setup environment.

On linux (arch at least) none of it works. the scripts run, however the UI is just white.

How i can access discord?, i'm not finding any links.

Sir pls look into https://github.com/QwenLM/Qwen3-TTS Its so fucking good and need just 3s audio. It can design, clone, vioce with tones and so on. Its better than elevenlabs (RIP) https://qwen.ai/blog?id=qwen3tts-0115 Swap that Sovits Server with a custom build server (fastapi) and use qwen3 tts

having problems with audio_playback

Update from a Mac user. Just keep throwing the errors and files into claude and eventually it worked. Have to start all 3 servers one by one but she’s alive.

Can someone help? I can’t run it cuz go-api.bat dosnt exist so i can’t run voice , is there a solution for it and /or can someone send it please or help?

mac user here, how does one connect the SOvits? the mac version seems to have many different files, can't seem to connect the dots on a mac. Was wondering if any mac users gave it a shot? Thanks!

Finaly fixed it! now, for everyone that has a rtx 50 series card and want to use it with GPT-SOVITS, you need a specific GPT-SOVITS version, not the one that can be download at the github repo., bc. of missmatching pytorch version for rtx 50 series. So that beening said, delete your currently GPT-SOVITS folder and just download this version from huggingface and extracted it. Then continue with it like in the video and boom -> It works :D https://huggingface.co/lj1995/GPT-SoVITS-windows-package/blob/main/GPT-SoVITS-v4-20250529-nvidia50.7z Dont forget to like his repo. on huggingface for this solution :) Also for the ones that has plenty VRAM on GPU, change the values in this file "tts_infer" to this, for faster responses (will load model completely in VRAM) \GPT_SoVITS\configs -> Find this file "tts_infer" and open it with your editor. device: cuda is_half: false

Ok found the error! :) It is again the RTX 50 Series that is causing a headache xD The problem is, that with the default GPT-SOVITS with api_v2 the "cuda" libs are not loading, bc. of missmatching versions. I need to find a way to integrated it natively with the one click version of GPT-SOVITS :D For the time being, i also found a workaround but it will definitely cook your CPU, and mine is a i7 13700k^^ Workaround: Navigate to this folder in your GPT-SOVITS Directory: 1. \GPT_SoVITS\configs -> Find this file "tts_infer" and open it with your editor. 2. Change the first lines to this: custom: bert_base_path: GPT_SoVITS/pretrained_models/chinese-roberta-wwm-ext-large cnhuhbert_base_path: GPT_SoVITS/pretrained_models/chinese-hubert-base device: cpu is_half: false What it does: In default it will take the cuda device (your GPU) to compile the txt2speech, but that will not work, like i said, so instead it will use your CPU for that, but keep in mind, it will load your cpu to 100% xD Will Update later with a "Cuda" solution :)

Got the same error :( Error in main loop: [Errno 2] No such file or directory: 'client\\audio\\output_32c3a876d4484519bba6bf00d8e4cd29.wav' It does also happen, when i try to execute the check.py. The strange thing is, that GPT-SOVITS will crash at decoding procedure, so maybe we need to set a output folder or something for temp decoding idk.

Error in sovits_gen: HTTPConnectionPool(host='127.0.0.1', port=9880): Max retries exceeded with url: /tts (Caused by NewConnectionError("HTTPConnection(host='127.0.0.1', port=9880): Failed to establish a new connection: [WinError 10061] No connection could be made because the target machine actively refused it")) Error in main loop: [Errno 2] No such file or directory: 'client\\audio\\output_32c3a876d4484519bba6bf00d8e4cd29.wav' I'm getting this whenever I try to initiate a conversation, any clue of what could be happening?

Question: if you are using almost everything locally, why don't you instead of using gpt API just make it use a soemthing local like ollama, why spend money on gpt when you can do it for free and it doesnt take anytime at all to generate reponse compared to generating voice, and not hard to switch your code!

start "GPT-SoVITS" cmd /k "cd /d %SOVITS_PATH% && call go-webui.bat" isnt valid anymore, when i start the API manual (api 1 or 2) the Audio output does not work for me - text gen / animation / voice rec is working, just no audio output (and no files generated in the /client/audio folder)

Hey, my sovits doesn't have a go-api ? I'm not sure if it's updated but even the github doesn't have the file listed. :')

just wanna ask if it is normal to have like 20 sec delay after i stop speaking to her reply?

this is infacct peak

Thank you for sharing the latest code. Literally this morning I was wondering if we might get an update soon. (What timing!) It's a fun project in parallel to learning python. After I get this update running happily, I think I may try to switch the TTS from SoVITS to Piper(?).. as I read that I might be able to leverage my AMD GPU for acceleration better than SOVITs... but we shall see (and of course there is Groq too that you have built in.. but I learn along the way if I succeed). Anyways, main message is: Thank you

Is this peak chat?

Great work, can you maybe one day add something so she can speak German? Sovits as far as I know is not capable of that.