Breaking Barriers for Minority Languages: Development and Evaluation of the First Galician On-Device ASR Model for Metaverse Applications

This paper presents the development and evaluation of the first Automatic Speech Recognition (ASR) model for the Galician language, specifically designed for Metaverse applications using Microsoft HoloLens 2 smart glasses. The Metaverse is an expanding digital space for immersive interactions using different Extended Reality (XR) technologies, where voice-based communication is essential for an intuitive User Experience (UX). For minority languages, voice interaction is especially valuable as it promotes accessibility in this new environment by removing linguistic barriers. While ASR systems for high-resource languages like English and Mandarin are becoming increasingly common in the Metaverse, there are no existing models optimized for Metaverse applications for Galician, a low-resource language spoken by approximately 3.2 million people. The developed ASR system leverages a fine-tuned Wav2Vec 2.0 model optimized specifically for on-device inference in ARM64 architectures, ensuring low latency without relying on cloud-based solutions. Performance evaluations indicate a Character Error Rate (CER) below 6.7 % and an average latency of 2.98 seconds, demonstrating the model effectiveness in real-time transcription tasks in computationally constrained devices such as the Microsoft HoloLens 2. Thus, this paper lays the groundwork for the future integration of Galician and other minority languages in the development of Metaverse applications.