[Bug] Voice cloning issue in coqui-tts 0.25.0

[mirfan899]

Describe the bug

I'm running tts on spanish and hindi language. The audio generated are better by previous repo. New repo seems to add female mixture to generated audio.

Here is the sample audio and the output generated by both repo.

I'm using the same code for both.

import torch
from TTS.api import TTS

# Get device
device = "cuda" if torch.cuda.is_available() else "cpu"

# Init TTS
tts = TTS("tts_models/multilingual/multi-dataset/xtts_v2").to(device)

tts.tts_to_file(text="After this, we will talk about a player who will give a lot of respect to Indian fans in this world cup. I will just tell you about this player who announced his entry in International Cricket and did a great performance. From today, we will meet you every day at EAM Cricket World Cup 2007.", speaker_wav="../audio/urdu.wav", language="en", file_path="output.wav")

data.zip

To Reproduce

Install old TTS and new TTS separately and test given code.

import torch
from TTS.api import TTS

# Get device
device = "cuda" if torch.cuda.is_available() else "cpu"

# Init TTS
tts = TTS("tts_models/multilingual/multi-dataset/xtts_v2").to(device)

tts.tts_to_file(text="After this, we will talk about a player who will give a lot of respect to Indian fans in this world cup. I will just tell you about this player who announced his entry in International Cricket and did a great performance. From today, we will meet you every day at EAM Cricket World Cup 2007.", speaker_wav="../audio/urdu.wav", language="en", file_path="output.wav")

Expected behavior

voice quality not good.

Logs

No response

Environment

{
    "CUDA": {
        "GPU": [
            "Tesla T4"
        ],
        "available": true,
        "version": "12.1"
    },
    "Packages": {
        "PyTorch_debug": false,
        "PyTorch_version": "2.5.1+cu121",
        "TTS": "0.22.0",
        "numpy": "1.22.0"
    },
    "System": {
        "OS": "Linux",
        "architecture": [
            "64bit",
            "ELF"
        ],
        "processor": "x86_64",
        "python": "3.10.12",
        "version": "#1 SMP PREEMPT_DYNAMIC Thu Jun 27 21:05:47 UTC 2024"
    }
}

Additional context

No response

[eginhard]

Thank you for the report! There is indeed an issue with voice cloning and I've identified the source. I should be able to release a fix in ~2 days. In the meantime you can use version 0.24.3 of this fork.

[eginhard]

Sorry about this. It should be back to normal now in version 0.25.1