README.md

metadata

language:
  - ku
size_categories:
  - 1K<n<10K
task_categories:
  - text-to-speech

KurFemTTS Dataset

Overview

This dataset contains audio chunks and their corresponding textual transcripts in Kurdish. It is designed for Text-to-Speech (TTS) research and development, specifically focusing on female Kurdish speech (as part of the KurFemTTS project).

Dataset Structure

The dataset consists of audio .wav files grouped into tracks and a single metadata file metadata.csv that provides the transcription for each audio file.

Metadata (metadata.csv)

The metadata.csv file contains 2,994 transcriptions. It has the following columns:

• file_name: The relative path to the .wav audio chunk (e.g., track1\track1_0001.wav).
• track: The source track or session number the chunk was extracted from.
• chunk_number: The sequential number of the chunk within the track.
• sentence: The Kurdish text transcription corresponding to the audio chunk.

Audio Data

• Format: WAV (.wav)
• Channels: Mono
• Total Duration: ~10 hours
• Speaker Profile: Female (Single speaker)
• The audio files are provided in a compressed .zip archive (e.g., wavs_2026.zip).
• Total number of audio chunks: 2,994.
• Total size of the zip archive: ~1.36 GB.

Usage

This dataset can be used to train or fine-tune Text-to-Speech (TTS) models or Automatic Speech Recognition (ASR) systems for the Kurdish language.

To load the dataset using the Hugging Face datasets library, you can parse the metadata.csv and load the associated .wav files.

Language

• Kurdish (Central Kurdish / Sorani)