Speech recognition mobile application

A Speech-to-Text model for a mobile app with automatic speech recognition for teaching children to read.

Since the app had to control children's pronunciation, we needed to implement algorithms for both directions that would provide high precision results. Existing solutions were not applicable, so we researched both available data sources and probable approaches to meet model restrictions.

We collected and augmented intensive data sources; designed custom architectures; implemented the solution, and tested it. We integrated several sound transfer techniques for the speech pattern reconstruction, created clusterization algorithms for the data segmentation and further analysis.