Samsung Research and Development Institute India-Bangalore (SRI-B), one of the South Korean tech giant's key innovation centers globally, worked with academic partners to develop the Hindi language for its Galaxy AI, the set of artificial intelligence (AI) for Samsung devices. The collaboration covered 20 different regional dialects of the language, tonal and colloquial inflections. One of the key partners, the Vellore Institute of Technology (VIT) helped source large amounts of data to train the large language model.
SRI-B develops Hindi language for Galaxy AI
In an editorial post on Samsung India's website, the company highlighted the role its R&D institute and academic partners have played in developing regional language understanding for Galaxy AI . The company said it is working to expand Galaxy AI into different global languages to allow more users to use these features in their native language. Language capability enhances some of the device's features, such as Live Translate, Interpreter, Note Assist, and Browsing Assist.
To develop the Hindi language for the AI suite, SRI-B worked together with VIT. The publication highlights that the academic institute helped obtain almost one million lines of segmented and selected audio data about speech, words and conversational commands.
The reason for the collaboration was the difficulty in generating high-quality data in Hindi due to the large amount of regional variance. The SRI-B team covered 20 different regional dialects, tonal inflections, punctuation and colloquialisms.
Samsung developed a facility for VIT under its Students Ecosystem for Engineered Data (SEED) Labs initiative in 2021. The company says the lab is equipped with head and torso simulators, binaural microphones and hearing devices surrounded by a advanced sound absorption system. It also calls for projects where university staff, students and interns can work alongside technical experts provided by the tech giant.
In particular, SRI-B collaborated with teams around the world to develop AI language models in British English, Indian, Australian English, Thai, Vietnamese and Indonesian. The role of the research center was considered critical and hence it was asked to develop the Hindi language for Galaxy AI.
