add kbes dataset

data/kbes/README.md

@@ -0,0 +1,13 @@
+# Nkululeko pre-processing for KBES dataset
+
+Download the dataset from [1], place it in this directory or somewhere else and
+ane make soft link here (`ln -sf`).
+
+```bash
+# unzipt the dataset
+unzip "KUET Bangla Emotional Speech (KBES) Dataset.zip"
+
+```
+
+Reference:  
+[1] <https://data.mendeley.com/datasets/vsn37ps3rx/4>

data/kbes/exp.ini

@@ -0,0 +1,29 @@
+[EXP]
+root = /tmp/results/
+name = exp_kbes_hubert_all
+[DATA]
+databases = ['train', 'dev', 'test']
+train = ./data/kbes/kbes_train.csv
+train.type = csv
+train.absolute_path = False
+train.split_strategy = train
+dev = ./data/kbes/kbes_dev.csv
+dev.type = csv
+dev.absolute_path = False
+dev.split_strategy = train
+test = ./data/kbes/kbes_test.csv
+test.type = csv
+test.absolute_path = False
+test.split_strategy = test
+target = emotion
+; labels = ['anger', 'fear', 'sad', 'happy']
+; get the number of classes from the target column automatically
+[FEATS]
+type = ['hubert-xlarge-ll60k']
+; no_reuse = False
+scale = standard
+[MODEL]
+type = knn
+; save = True
+[RESAMPLE]
+replace = True

data/kbes/process_database.py

@@ -0,0 +1,99 @@
+# process_database.py for KBES dataset
+
+import pandas as pd
+import argparse
+from nkululeko.utils.files import find_files
+import os
+from sklearn.model_selection import train_test_split
+
+
+def process_database(data_dir, output_dir):
+    # check if data_dir exist
+    if not os.path.exists(data_dir):
+        raise FileNotFoundError(f"Directory {data_dir} not found.")
+
+    # Create the output directory if it doesn't exist
+    os.makedirs(output_dir, exist_ok=True)
+
+    # read all wav files
+    wavs = find_files(data_dir, relative=True, ext=["wav"])
+    print(f"Found {len(wavs)} wav files.")
+
+    # map emotion: 1 = Neutral, 2 = Happy, 3 = Sad, 4 = Angry, 5 = Disgust
+    emotion_mapping = {
+        1: "neutral",
+        2: "happy",
+        3: "sad",
+        4: "angry",
+        5: "disgust"
+    }
+
+    # map intensity, 1 = low, 2 = high
+    intensity_mapping = {
+        1: 'low',
+        2: 'high'
+    }
+
+    # map gender 0 = female, 1 = male
+    gender_mapping = {
+        0: 'female',
+        1: 'male'
+    }
+
+    data = []
+    for wav in wavs:
+        # get basename
+        basename = os.path.basename(wav)
+        # get emotion
+        emotion = emotion_mapping[int(basename.split("-")[0])]
+        # get intensity
+        intensity = intensity_mapping[int(basename.split("-")[1])]
+        # get gender
+        gender = gender_mapping[int(basename.split("-")[2])]
+        # add language
+        language = "bangla"
+        # add to data list
+        data.append({
+            "file": wav,
+            "emotion": emotion,
+            "gender": gender,
+            "intensity": intensity,
+            "language": language
+        })
+
+    # create dataframe from data
+    df = pd.DataFrame(data)
+    # split the data into train, dev, and test sets, balanced by emotion
+    train_df, temp_df = train_test_split(
+        df, test_size=0.3, stratify=df['emotion'], random_state=42)
+    dev_df, test_df = train_test_split(
+        temp_df, test_size=0.5, stratify=temp_df['emotion'], random_state=42)
+    # write dataframes to csv
+    train_df.to_csv(os.path.join(
+        output_dir, "kbes_train.csv"), index=False)
+    dev_df.to_csv(os.path.join(output_dir, "kbes_dev.csv"), index=False)
+    test_df.to_csv(os.path.join(
+        output_dir, "kbes_test.csv"), index=False)
+    print(f"Number of train samples: {len(train_df)}")
+    print(f"Number of dev samples: {len(dev_df)}")
+    print(f"Number of test samples: {len(test_df)}")
+    print("Database processing completed.")
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description="Process KBES dataset")
+    parser.add_argument(
+        "--data_dir",
+        type=str,
+        default="KUET Bangla Emotional Speech (KBES) Dataset",
+        help="Directory containing the KBES data",
+    )
+    parser.add_argument(
+        "--output_dir",
+        type=str,
+        default="./",
+        help="Directory to store the output CSV files",
+    )
+    args = parser.parse_args()
+
+    process_database(args.data_dir, args.output_dir)

data/nemo/exp.ini

@@ -1,6 +1,6 @@
 [EXP]
 root = /tmp/results/
-name = exp_nemo_audmodel_svm_big4
+name = exp_nemo_audmodel_knn
 [DATA]
 databases = ['train', 'dev', 'test']
 train = ./data/nemo/nemo_train.csv
@@ -19,7 +19,7 @@ test.absolute_path = False
 test.split_strategy = test
 test.audio_path = ./nEMO/samples
 target = emotion
-labels = ['anger', 'neutral', 'sadness', 'happiness']
+; labels = ['anger', 'neutral', 'sadness', 'happiness']
 ; get the number of classes from the target column automatically
 [FEATS]
 type = ['audmodel']

nkululeko/data/dataset_csv.py

@@ -21,7 +21,7 @@ def load(self):
         #     exp_root = self.util.config_val("EXP", "root", "")
         #     data_file = os.path.join(exp_root, data_file)
         root = os.path.dirname(data_file)
-        audio_path = self.util.config_val_data(self.name, "audio_path", "")
+        audio_path = self.util.config_val_data(self.name, "audio_path", "./")
         df = pd.read_csv(data_file)
         # special treatment for segmented dataframes with only one column:
         if "start" in df.columns and len(df.columns) == 4:
@@ -49,7 +49,8 @@ def load(self):
                     .map(lambda x: root + "/" + audio_path + "/" + x)
                     .values
                 )
-                df = df.set_index(df.index.set_levels(file_index, level="file"))
+                df = df.set_index(df.index.set_levels(
+                    file_index, level="file"))
             else:
                 if not isinstance(df, pd.DataFrame):
                     df = pd.DataFrame(df)
@@ -63,7 +64,8 @@ def load(self):
         self.db = None
         self.got_target = True
         self.is_labeled = self.got_target
-        self.start_fresh = eval(self.util.config_val("DATA", "no_reuse", "False"))
+        self.start_fresh = eval(
+            self.util.config_val("DATA", "no_reuse", "False"))
         is_index = False
         try:
             if self.is_labeled and not "class_label" in self.df.columns:
@@ -90,7 +92,8 @@ def load(self):
                 f" {self.got_gender}, got age: {self.got_age}"
             )
         self.util.debug(r_string)
-        glob_conf.report.add_item(ReportItem("Data", "Loaded report", r_string))
+        glob_conf.report.add_item(ReportItem(
+            "Data", "Loaded report", r_string))
 
     def prepare(self):
         super().prepare()