A Speech Corpus for Mizo Automatic Speech Recognition: Whisper and SraVaani 1.0 Fine-Tuning with Morphology-Aware Evaluation
Researchers have developed a new Automatic Speech Recognition (ASR) system for the Mizo language. To improve the system's accuracy, they collected and fine-tuned it with three multilingual models: Whisper-large-v3, which achieved a low WER of 18.08%, and SraVaani 1.0 Indic, which improved from 58.27% to 17.93% after fine-tuning. The study's results show that the Whisper model can adapt well to an unseen language like Mizo.
Researchers have developed a new Automatic Speech Recognition (ASR) system for the Mizo language. To improve the system's accuracy, they collected and fine-tuned it with three multilingual models: Whisper-large-v3, which achieved a low WER of 18.08%, and SraVaani 1.0 Indic, which improved from 58.27% to 17.93% after fine-tuning. The study's results show that the Whisper model can adapt well to an unseen language like Mizo.
---
Why it matters: This research matters because it enables more accurate speech recognition in low-resource languages like Mizo, which will improve accessibility for speakers of these languages and facilitate further development of ASR systems in similar languages.
Source: https://arxiv.org/abs/2608.19361
This article was originally published at: https://arxiv.org/abs/2608.19361