ENV.py

SAMPLE_RATE = 16000
REPRESENTATION = "mert" # raw, mel, mfcc, melody, lyrics, raw+lyrics, mert

NEPTUNE_SWITCH = 1
Evaluate_Frequency = 1
### No need to change this part
### unless you want to change the dataset name or for other custom purpose
# the original dataset name
# Root_PATH = "/nfs/guille/eecs_research/soundbendor/shengxuan/opera2324/opera-singing-dataset/"
Root_PATH = ""
Data_PATH = Root_PATH + "operadataset2023/"
Unified_PATH = Root_PATH + 'unified/'
Trimmed_PATH = Root_PATH + 'trimmed'

### hyperparameters related to Machine Learning
### keep them consistent during EVERY workflow steps (from data preprocessing to evaluation)
target_second = 30
segment_method = "Padding-S" # "Padding-S(ilence)" or "Padding-C(ircular)" or "Dropping"
target_class = "emotion_binary"
fold_count = 5

target_class_dictionary = {
    "emotion_binary": "Emo",
    "bio_gender": "Gen",
    "level": "Lev",
    "role": "Rol",
    "acappella": "Aca",
    "jingju": "Jin",
}

target_class_short = target_class_dictionary[target_class]

### !!! MUST NOT Change this part !!!
Trimmed_PATH = Trimmed_PATH + "_" + str(target_second) + "_" +segment_method
### !!! MUST NOT Change this part !!!

## melody extraction part
pretrained_models = "/nfs/guille/eecs_research/soundbendor/models/Melody-extraction-with-melodic-segnet/pretrain_model/"