Dhivehi-Audio-Casts

Audio dataset with transcripts and speaker characteristics extracted from Qdrant database.

Dataset Summary

This dataset contains 3,300 audio recordings with corresponding transcripts and speaker characteristics including:

• Audio: WAV format audio files
• Sentence: Transcribed text in Dhivehi
• Speaker Demographics: Age, gender probabilities (male/female/child)
• Audio Characteristics: Arousal, dominance, valence scores
• Audio Length: Duration in seconds

Dataset Structure

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("alakxender/dhivehi-audio-casts", use_auth_token=True)

# Example record
{
    'audio': {
        'path': 'batch340_0_speaker_00_part5_original.wav',
        'array': [...],  # Audio array
        'sampling_rate': 16000
    },
    'sentence': 'އޭންގެ އުމުރުފުޅުން ސައްބީސް އަހަރު...',
    'audio_length': 29.24,
    'age': 27,
    'male': 0.995,
    'female': 0.004,
    'child': 0.0002,
    'arousal': 0.354,
    'dominance': 0.406,
    'valence': 0.350
}

Usage

This dataset can be used for:

• Automatic Speech Recognition (ASR) in Dhivehi testing
• Text-to-Speech (TTS) synthesis testing
• Speaker analysis and characterization
• Audio classification tasks

Data Fields

• audio: Audio file with path, array, and sampling rate
• sentence: Transcribed text in Dhivehi (Thaana script)
• audio_length: Duration of audio in seconds
• age: Speaker age
• male: Probability of male speaker (0-1)
• female: Probability of female speaker (0-1)
• child: Probability of child speaker (0-1)
• arousal: Audio arousal score (0-1)
• dominance: Audio dominance score (0-1)
• valence: Audio valence score (0-1)

Language

The dataset is in Dhivehi (Maldivian), written in Thaana script.