TAL Education Releases 587‑Hour Bilingual Speech Dataset for AI Research

TAL Education, a pioneer in education technology, recently released a 587‑hour bilingual (Chinese‑English) speech dataset collected from teacher English‑instruction classroom scenes, making it one of the largest open speech datasets in the education sector and the largest known mixed‑language dataset globally.

The dataset addresses the critical shortage of high‑quality mixed‑language speech data for artificial intelligence research, where algorithm, computing power, and data are the three pillars of AI development. High‑quality data can significantly improve model training and prediction accuracy.

As a leading AI‑focused education company, TAL Education has long invested in AI applications for education, accumulating extensive teaching resources and massive data. It also contributes to the national "Smart Education Next‑Generation AI Open Innovation Platform," committing to accelerate resource sharing and technological innovation through open sourcing.

Since March 2020, TAL Education has released several open datasets, including a primary school arithmetic dataset, handwritten Chinese‑English text, handwritten formulas, Chinese speech recognition, and speech emotion datasets. Notably, its handwritten formula dataset served as the official data for the 5th China Innovation Challenge "Education Handwritten Formula Recognition" competition, attracting top universities and leading internet companies.

With 18 years of rapid growth, TAL Education has built cross‑business technical integration mechanisms and a technology middle‑platform, laying a solid foundation for open sourcing its educational data. Ongoing development of the national AI open platform will further expand the availability of such datasets, fostering a collaborative ecosystem for smart education.

Dataset download instructions: click "Read Original" to view details, copy the link https://ai.100tal.com/dataset, and open it on a computer to download.