INUR TECHNOLOGIES LLC — Department of Defense SBIR Phase I: SOCOM232-002

Description

Requirement of a portable device that can translate spoken conversations between Taiwan Hokkien and English. No commercially available product exists that can provide this capability, which is 100% disconnected from the cloud. The first challenge is to ensure the device's portability and accuracy of the translation without an Internet connection. The second challenge is the lack of existing data to train Machine Learning models due to Taiwan Hokkien being primarily a spoken language. Our approach is the systematic collection and annotation of Hokkien speech data to ensure maximum translation accuracy. Two data collection methods will be employed, including a web-based system with QR code cards distributed to volunteers in Taiwan and an investigator interviewing local subjects to fill gaps in data. The data collection strategy aims to capture a broad spectrum of idioms from various age groups and linguistic features like Sandhi and tonal variations. After data collection, ML models will be employed on single-board computers to develop a prototype that meets the design parameters of complete disconnection from the Internet, and having speed, and accuracy of the translation.