VitsModelSplit/finetune_config_ara.json

{
    "project_name": "vits_ara",
    "push_to_hub": false,
    "hub_model_id": "ara_tts_finetuning/ara_tts_finetuning",
    "overwrite_output_dir": true,
    "output_dir": "./output",

    "dataset_name": "./dataset/",      
    "dataset_config_name": "welsh_female", 
    "audio_column_name": "audio", 
    "text_column_name":"text",
    "train_split_name": "train",
    "eval_split_name": "eval",
  
    "override_speaker_embeddings": false,
    "filter_on_speaker_id": 5223,

    
    "max_duration_in_seconds": 20,
    "min_duration_in_seconds": 1.0,
    "max_tokens_length": 500,

    "model_name_or_path": "facebook/mms-tts-ara",

    "full_generation_sample_text": "اوريه و اخليه يعرف هو حاط نفسه في مواجهة مع مين",
    "preprocessing_num_workers": 4,

    "do_train": true,
    "num_train_epochs": 300,
    "gradient_accumulation_steps": 1,
    "per_device_train_batch_size": 10,
    "learning_rate": 2e-5,
    "adam_beta1": 0.8,
    "adam_beta2": 0.99,
    "warmup_ratio": 0.01,
    "d_learning_rate": 2e-5,
    "d_adam_beta1": 0.7,
    "d_adam_beta2": 0.99,

    "do_eval": true, 
    "eval_steps": 10,
    "per_device_eval_batch_size": 10,
    "max_eval_samples": 2, 
    "do_step_schedule_per_epoch": true,

    "weight_disc": 3,
    "weight_fmaps": 1,
    "weight_gen": 1,
    "weight_kl": 1.5,
    "weight_duration": 1,
    "weight_mel": 35,

    "fp16": false,
    "seed": 456
}