xttsv2 model sometimes（almost 10%）produce extra noise.[Bug]

[seetimee]

Describe the bug

For example,If it generates a 30-second audio, for the first 15 seconds it generates normal audio, and for the last 15 seconds it generates noise

To Reproduce

tts_to_file(text=text,
language=lang_code,
speaker_wav=speaker_wav,
speed=speed,
file_path=temp_file
,split_sentences=split_sentences)

Expected behavior

No response

Logs

No response

Environment

{
    "CUDA": {
        "GPU": [
            "NVIDIA GeForce RTX 3080"
        ],
        "available": true,
        "version": "11.8"
    },
    "Packages": {
        "PyTorch_debug": false,
        "PyTorch_version": "2.1.2+cu118",
        "TTS": "0.22.0",
        "numpy": "1.23.0"
    },
    "System": {
        "OS": "Linux",
        "architecture": [
            "64bit",
            "ELF"
        ],
        "processor": "x86_64",
        "python": "3.10.8",
        "version": "#148-Ubuntu SMP Mon Oct 17 16:02:06 UTC 2022"
    }
}

Additional context

No response

[TangHaitao1994]

training data default length is about 15 seconds

[seetimee]

Is my training reference audio too long?（lower than 5 minutes will be better）

[kaveenkumar]

Most often occurs in FR language than others.

You synthesize 10 samples of a "text" for a given "voice", 2-3 out of those 10 samples contain the "text" + "some other blabbering"

[Sharrnah]

same with german.
English seems fine, so my guess is the training for some languages resulted in these issues.

For german, i think i sometimes could hear sentences like Punkt. Neues Kapitel or similar.
So maybe the training data included some verbose audio the readers added and the AI "learned" this for these languages.

(just some guessing here)

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions. You might also look our discussion channels.

[seetimee]

keep watching

[Sharrnah]

i think something that confirms my guess.
In Italian, if you add a . (dot) at the end of a sentence, it very often speaks that dot "punkto".

it should not pronounce punctuation marks in my opinion.

[CRochaVox]

I also have a problem with noise after punctuations and in long sentences it is common for him to say the 'ponto' for the period in Portuguese

Does anyone know if it is possible to reduce this by finetunning the model?

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions. You might also look our discussion channels.