Elyordev/combined_dataset

Combined Dataset

Uzbek Speech Recognition Dataset

Description

This dataset is developed for training and evaluating speech recognition models in the Uzbek language. It includes audio files, their corresponding transcriptions, and metadata like speaker demographics (age, gender, accent). The dataset is designed to be fully compatible with the Hugging Face Dataset Viewer, allowing users to explore audio files directly in the browser.

Dataset Structure

The dataset is organized into the following folders:

combined_dataset/
├── audio/
│   ├── train/
│   ├── test/
│   └── validation/
├── transcript/
│   ├── train.tsv
│   ├── test.tsv
│   └── validation.tsv
├── .gitattributes
├── README.md
└── release_stats.py

TSV File Details

Each .tsv file contains the following columns:

• id: Unique identifier for the audio file.
• path: Relative path to the audio file.
• sentence: Transcription of the audio.
• duration: Duration of the audio in seconds.
• age: Age group of the speaker (e.g., 18-24, 25-34).
• gender: Gender of the speaker (e.g., Erkak, Ayol).
• accents: Dialect or accent of the speaker
• locale: Language of the speaker (e.g., O'zbek tili).

Features

• Audio Playback: Audio files can be played directly in the Hugging Face Dataset Viewer.
• Rich Metadata: Includes speaker information (age, gender, accent).
• Language: All transcriptions are in Uzbek.

How to Use the Dataset

Using Python

1. Install the datasets library:

   pip install datasets
   

2. Load the dataset:

   from datasets import load_dataset
   
   dataset = load_dataset("username/combined_dataset")
   

3. Display a sample:

   print(dataset["train"][0])
   

4. Listen to an audio file:

   from IPython.display import Audio
   
   Audio(dataset["train"][0]["path"])
   

Citation

@dataset{uzbek_speech_recognition,
  author = {Elyor},
  title = {Combined_dataset (uzbekvoice, commonvocie, issai datasets)},
  year = {2025},
  url = {https://huggingface.co/datasets/username/combined_dataset}
}