Uzbek Speech Transcription Dataset

This dataset contains Uzbek speech audio chunks with their transcriptions. The audio files are automatically segmented using VAD (Voice Activity Detection).

Dataset Structure

Each entry contains:

audio: Audio file (FLAC format, 24kHz sampling rate)
text: Uzbek transcription text
speaker_gender: Estimated speaker gender
noise_level: Background noise level assessment
contains_music: Whether background music is detected
contains_overlap: Whether speaker overlap is detected
speakers_estimate: Estimated number of speakers
confidence: Transcription confidence score
is_uzbek_only: Whether the audio contains only Uzbek speech
has_unintelligible_speech: Whether the audio contains unintelligible speech
has_audio_cuts: Whether the audio has abrupt cuts or interruptions
filename: Original filename
upload_timestamp: When this entry was uploaded

Usage

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("capsul1912/Ilmparvar_samples")

# Access audio and text
for example in dataset["train"]:
    audio = example["audio"]
    text = example["text"]
    print(f"Text: {text}")

Total entries: 1677 Last updated: 2025-07-27 21:44:38