Skip to content

Dataset · cloud0day3

3411 hours of synthetic Turkish speech with voice descriptions

cloud0day3/alania-synthetic-speech-tr

Over 2M Turkish audio clips at 48 kHz with transcripts and timbre descriptions for training TTS and ASR.

What's inside

A massive corpus of 2,037,461 synthetic audio clips (3,411 hours) sampled at 48 kHz in Turkish. The recordings are voiced by 2,752 generated voices without using real human voices, and each includes the original text, phonetic transcription, and an English voice description. The dataset was created during the development of the Alania-2 Turkish speech synthesis model.

What you can do with it

  • Training Turkish text-to-speech (TTS) models with voice control via text descriptions
  • Training and augmenting Turkish automatic speech recognition (ASR) models
Tasks
Text → speech, Automatic speech recognition
Languages
Turkish
Size
1M–10M rows
Downloads this month
5.1K
Likes
52

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("cloud0day3/alania-synthetic-speech-tr")