[Bug] xtts voice generation is better in 0.24.3 than in 0.25 and above

[C00reNUT]

Describe the bug

Hello, thank you for maintaining this library, this is probably related to #198 - when i am using 0.24.3 version for inference of xtts model I get much better results than in 0.25.1 - there must be still some bug in the inference. I didn't try the exactly same generation with the same seed, but the quality difference is obvious.

I would provide some samples but I am using czech finetuned model and you couldn't really hear the difference unless you are native...

To Reproduce

import os
import torch
import torchaudio
from TTS.tts.configs.xtts_config import XttsConfig
from TTS.tts.models.xtts import Xtts

print("Loading model...")
config = XttsConfig()
config.load_json("/path/to/xtts/config.json")
model = Xtts.init_from_config(config)
model.load_checkpoint(config, checkpoint_dir="/path/to/xtts/", use_deepspeed=True)
model.cuda()

print("Computing speaker latents...")
gpt_cond_latent, speaker_embedding = model.get_conditioning_latents(audio_path=["reference.wav"])

print("Inference...")
out = model.inference(
"It took me quite a long time to develop a voice and now that I have it I am not going to be silent.",
"en",
gpt_cond_latent,
speaker_embedding,
temperature=0.7, # Add custom parameters here
)
torchaudio.save("xtts.wav", torch.tensor(out["wav"]).unsqueeze(0), 24000)

Expected behavior

The outputs shall be similar using same parameters, accounting for diffusion variability in outputs.

Logs

No response

Environment

{
    "CUDA": {
        "GPU": [
            "NVIDIA GeForce RTX 3090"
        ],
        "available": true,
        "version": "12.4"
    },
    "Packages": {
        "PyTorch_debug": false,
        "PyTorch_version": "2.5.1",
        "TTS": "0.24.3",
        "numpy": "1.26.4"
    },
    "System": {
        "OS": "Linux",
        "architecture": [
            "64bit",
            "ELF"
        ],
        "processor": "x86_64",
        "python": "3.11.9",
        "version": "#49-Ubuntu SMP PREEMPT_DYNAMIC Mon Nov  4 02:06:24 UTC 2024"
    }
}

Additional context

No response

[eginhard]

I just double-checked for some examples that 0.25.1 and 0.24.3 (and other previous versions) produce exactly the same output when fixing the seed. If you could share some samples and/or test with a fixed seed that would be helpful.

[C00reNUT]

I tried it once again and I was using wrong config for selected model version, sorry about the confusion, happy new year!