Assamese Speech Labeled Dataset

The Assamese Speech Emotion Dataset  is a labeled speech corpus comprising 500 audio samples in the Assamese language, specifically designed for speech emotion recognition tasks. It features recordings from 10 male native speakers, each uttering 5 distinct sentences with 5 different emotions—Neutral, Angry, Happy, Fear, and Sad. For every sentence-emotion pair, each speaker recorded two utterances, resulting in consistent and well-distributed emotional speech data.

Directory structure :

The dataset is organized by speaker, with each speaker's recordings stored in a separate folder named Actor1 through Actor10. Inside each folder, there are 50 .wav files corresponding to all combinations of 5 sentences, 5 emotions, and 2 repetitions per pair.

where each file is named using the following format<sentenceNumber><emotionLabel><repetitionNumber>.wav

For example, 2happy1.wav in folder Actor1 represents the first repitition of second sentence spoken with happy emotion.

Direction of Use:

To extract the label from the dataset one can use the following code sample :

labels=[]

paths=[]

base_path=base path of the dataset folder

for actor_folder in os.listdir(base_path):            actor_path = os.path.join(base_path, actor_folder)            if os.path.isdir(actor_path):                for filename in os.listdir(actor_path):                    file_path = os.path.join(actor_path, filename)                    paths.append(file_path)                    base_name = filename.split('.')[0]                    label = ''.join([char for char in base_name if char.isalpha()])                    labels.append(label.lower())