Dataset · ysdede
Turkish speech corpus from Khan Academy videos
ysdede/khanacademy-turkish
78 hours of short audio clips with Turkish voiceovers and text for ASR and TTS training.
What's inside
Contains short audio fragments with an average duration of 10.5 seconds along with corresponding Turkish texts. Total recording volume is 78 hours (10 to 100 thousand examples), with the source video directory tree attached. Data was extracted from the Turkish YouTube channel of the Khan Academy educational project and preprocessed.
What you can do with it
- Training and fine-tuning Turkish speech recognition (ASR) models
- Training Turkish text-to-audio synthesis models
- Tasks
- Automatic speech recognition, Text → audio
- Languages
- Turkish
- Size
- 10K–100K rows
- Downloads this month
- 270
- Likes
- 39
Models trained on it
- mstkyvz/embeddinggemma2-turkish-sttAutomatic speech recognition 53
- ysdede/Phi-4-mm-inst-asr-turkish-unfText generation 44
- thealper2/whisper-base-turkish-khanacademyAutomatic speech recognition 35
- khazarai/KhanAcademy-TTSText → speech 34
- berrinuzun/whisper-large-v3-turkish 17
- kadirnar/Orpheus-KhanAcademy-TRText generation 10
- atasoglu/seda-v0.2Automatic speech recognition 0
How to load it
from datasets import load_dataset
ds = load_dataset("ysdede/khanacademy-turkish")