AssertionError: [!] You do not have enough samples for the evaluation set. You can work around this setting the 'eval_split_size' parameter to a minimum of 0.029411764705882353

[PankidT]

I need some help. I want to do voice cloning project with coqui-tts with using vits model. I got this bug in step of start training. This is my config

audio_config = VitsAudioConfig(
    sample_rate=22050, win_length=1024, hop_length=256, num_mels=80, mel_fmin=0, mel_fmax=None
)

config = VitsConfig(
    audio=audio_config,
    run_name="vits_ljspeech_ly",
    batch_size=16,
    eval_batch_size=16,
    batch_group_size=5,
    num_loader_workers=4,
    num_eval_loader_workers=4,
    run_eval=True,
    test_delay_epochs=-1,
    epochs=100000,
    save_step=1000,
	save_checkpoints=True,
	save_n_checkpoints=4,
	save_best_after=2000,
    text_cleaner="english_cleaners",
    use_phonemes=True,
    phoneme_language="en-us",
    phoneme_cache_path=os.path.join(output_path, "phoneme_cache"),
    compute_input_seq_cache=True,
    print_step=25,
    print_eval=True,
    mixed_precision=True,
    output_path=output_path,
    datasets=[dataset_config],
    cudnn_benchmark=False,
)

And for sample

train_samples, eval_samples = load_tts_samples(
    dataset_config,
    eval_split=True,
    eval_split_max_size=config.eval_split_max_size,
    eval_split_size=config.eval_split_size,
)

[fkwlqm]

just use a number bigger than 1 (eg. 12) to use that number as your eval data.

[PankidT]

It works, thank you very much.

[Vishweshwar133]

Where to write this

[fkwlqm]

in the config , write eval_split_size = 10 (or whatever number of files you need).