add eryca dataset

data/erysac/README.md

@@ -0,0 +1,21 @@
+# Nkululeko pre-processing for EmoChildRu Dataset
+
+```bash
+# grad the dataset from github
+git clone https://github.com/hydra-colab/Emotion-Recognition-of-Younger-School-Age-Children
+
+# rename WAV with wav
+find Emotion-Recognition-of-Younger-School-Age-Children -name '*.WAV' -exec bash -c 'f="{}"; mv -- "$f" "${f%.WAV}.wav"' \;
+# check number of file, should be 2505
+find Emotion-Recognition-of-Younger-School-Age-Children/ -name '*.wav' | wc -l
+# process database
+python3 process_database.py
+# resample and run
+cd ../..
+python3 -m nkululeko.resample --config data/erysac/exp.ini
+python3 -m nkululeko.nkululeko --config data/erysac/exp.ini
+```
+
+
+Reference:  
+[1] E. Lyakso et al., “EmoChildRu: Emotional child Russian speech corpus,” Lect. Notes Comput. Sci. (including Subser. Lect. Notes Artif. Intell. Lect. Notes Bioinformatics), vol. 9319, pp. 144–152, 2015, doi: 10.1007/978-3-319-23132-7_18.

data/erysac/exp.ini

@@ -0,0 +1,30 @@
+[EXP]
+root = ./results/
+name = exp_erysac_wavlm_os_knn
+[DATA]
+databases = ['train', 'dev', 'test']
+train = ./data/erysac/erysac_train.csv
+train.type = csv
+train.absolute_path = False
+train.split_strategy = train
+dev = ./data/erysac/erysac_dev.csv
+dev.type = csv
+dev.absolute_path = False
+dev.split_strategy = train
+test = ./data/erysac/erysac_test.csv
+test.type = csv
+test.absolute_path = False
+test.split_strategy = test
+target = emotion
+; labels = ['anger', 'neutral', 'fear']
+; get the number of classes from the target column automatically
+[FEATS]
+; type = ['audmodel']
+; type = ['hubert-large-ll60k']
+type = ['wavlm-large']
+scale = standard
+[MODEL]
+type = knn
+; save = True
+[RESAMPLE]
+replace = True

data/erysac/process_database.py

@@ -0,0 +1,68 @@
+# process_database.py for emochildru datasets
+
+import argparse
+from pathlib import Path
+from nkululeko.utils.files import find_files
+import pandas as pd
+
+gender_map = {"m": "male", "f": "female"}
+
+def process_database(data_dir, output_dir):
+    # check if data_dir exists
+    data_dir = Path(data_dir)
+    if not data_dir.exists():
+        raise FileNotFoundError(f"Directory {data_dir} not found")
+
+    # check if output_dir exists, create if not
+    output_dir = Path(output_dir)
+    if not output_dir.exists():
+        output_dir.mkdir(parents=True)
+
+    # find all wav files 
+    wavs = find_files(data_dir, ext=["wav"], relative=True)
+
+    # extract filename as df
+    df = pd.DataFrame({"file": wavs})
+
+    # extract emotion from filename's parent
+    df["emotion"] = df["file"].apply(lambda x: Path(x).parent.name.lower())
+
+    # extract speaker first string before _ in the filename
+    df["speaker"] = df["file"].apply(lambda x: Path(x).stem.split("_")[0])
+
+    # extract gender, second string, m for male f for female, map
+    df["gender"] = df["file"].apply(lambda x: Path(x).stem.split("_")[1])
+    # map m to male, f to female
+    df["gender"] = df["gender"].apply(lambda x: gender_map[x])
+
+    # extract age, third string, remove the suffix y
+    df["age"] = df["file"].apply(lambda x: Path(x).stem.split("_")[2][:-1])
+
+    # check number of speakers
+    speakers = df["speaker"].unique()
+    print(f"Number of speakers: {len(speakers)}")
+
+    # allocate 20% of speakers as test
+    test_speakers = speakers[: len(speakers) // 5]
+    df_test = df[df["speaker"].isin(test_speakers)]
+
+    # allocate 20% train for dev
+    dev_speakers = speakers[len(speakers) // 5 : len(speakers) // 5 * 2]
+    df_dev = df[df["speaker"].isin(dev_speakers)]
+    df_train = df.drop(df_dev.index)
+
+    # save to CSV
+    for split in ["train", "dev", "test"]:
+        df_split = eval(f"df_{split}")
+        df_split.to_csv(output_dir / f"erysac_{split}.csv", index=False)
+        print(f"Saved {split} set to {output_dir / f'erysac_{split}.csv'}"
+              f"with {len(df_split)} samples")
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--data_dir", type=str, default="./Emotion-Recognition-of-Younger-School-Age-Children/")
+    parser.add_argument("--output_dir", type=str, default="./")
+    args = parser.parse_args()
+
+    process_database(args.data_dir, args.output_dir)

data/turev/exp.ini

@@ -1,6 +1,6 @@
 [EXP]
 root = ./results/
-name = exp_turev_os_knn
+name = exp_wavlm_os_knn
 [DATA]
 databases = ['train', 'dev', 'test']
 train = ./data/turev/turev_train.csv
@@ -19,10 +19,10 @@ target = emotion
 ; labels = ['anger', 'neutral', 'fear']
 ; get the number of classes from the target column automatically
 [FEATS]
-type = ['os']
-; type = ['hubert-xlarge-ll60k']
-; no_reuse = False
-; scale = standard
+; type = ['audmodel']
+; type = ['hubert-large-ll60k']
+type = ['wavlm-large']
+scale = standard
 [MODEL]
 type = knn
 ; save = True