[Bug] eval_split_size in recipe and config.json don't match up

[fijipants]

🐛 Description

Preface: I'm not sure if this is expected behaviour or not, I'm just guessing here.

Scenario 1:

Let's say I want to run a training with a custom eval_split_size. I see that eval_split_size is in BaseTTSConfig, so I modify this line:

TTS/recipes/ljspeech/vits_tts/train_vits.py

Line 34 in 0cf3265

config = VitsConfig(

to this:

config = VitsConfig(eval_split_size=0.05,

Result: The eval_split_size does not get set properly. (It's set to 0.01 instead over here instead.)

Scenario 2:

Let's say I want to run a training with a custom eval_split_size, and I know that editing the base config won't work. Instead I modify this line:

TTS/recipes/ljspeech/vits_tts/train_vits.py

Line 72 in 0cf3265

train_samples, eval_samples = load_tts_samples(dataset_config, eval_split=True)

To this:

train_samples, eval_samples = load_tts_samples(dataset_config, eval_split=True, eval_split_size=0.05)

Result: Although it works, the outputted config.json has the wrong eval_split_size:

    "eval_split_size": 0.01,

Scenario 3:

Let's say I completed a training with a custom eval_split_size, and now I want to change the eval_split_size and continue it, so I open up the outputted config.json and change this line:

    "eval_split_size": 0.01,

To this:

    "eval_split_size": 0.07,

Then I start the training with --continue_path.

Result: It uses whichever eval_split_size is set on this line, instead of the one from config.json.

To Reproduce

See description.

Expected behavior

1. I should be able to set it with config = VitsConfig(eval_split_size=0.05,.
2. If that's not possible, then when I set it with load_tts_samples(..., eval_split_size=0.05), I'd like it to be outputted to the config.json.
3. When I continue a run, it should re-use the eval_split_size from the config.json.

Environment

{
    "CUDA": {
        "GPU": [
            "NVIDIA GeForce RTX 3090",
            "NVIDIA GeForce RTX 3090"
        ],
        "available": true,
        "version": "10.2"
    },
    "Packages": {
        "PyTorch_debug": false,
        "PyTorch_version": "1.11.0+cu102",
        "TTS": "0.6.1",
        "numpy": "1.19.5"
    },
    "System": {
        "OS": "Linux",
        "architecture": [
            "64bit",
            ""
        ],
        "processor": "x86_64",
        "python": "3.7.11",
        "version": "#202202230823 SMP PREEMPT Wed Feb 23 14:53:24 UTC 2022"
    }
}

Additional context

[Edresson]

@fijipants Thanks for report the bug. Indeed in #1251 we forget to change it for all the recipes. The PR #1424 will fix it :).