Summary of Whisper Finetuning on Nepali Language, by Sanjay Rijal et al.
Whisper Finetuning on Nepali Language
by Sanjay Rijal, Shital Adhikari, Manish Dahal, Manish Awale, Vaghawan Ojha
First submitted to arxiv on: 19 Nov 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper tackles the challenge of developing robust Automatic Speech Recognition (ASR) models for underrepresented languages like Nepali. The authors create a comprehensive dataset and fine-tune OpenAI’s Whisper models of varying sizes to improve transcription accuracy. They leverage publicly available datasets, self-recorded custom data, and augmentation techniques to enhance model robustness. Experimental results show that fine-tuning on the curated custom dataset reduces Word Error Rate (WER) across all model sizes due to variations in speaker characteristics, acoustic environments, dialects, and manual curation. The approach outperforms Whisper’s baseline models, achieving WER reductions of up to 36.2% on small and 23.8% on medium models. Data augmentation plays a crucial role in enhancing model robustness. This research highlights the importance of dataset quality, variation, and augmentation for adapting state-of-the-art models to underrepresented languages. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps make better computers that can understand Nepali speech. Right now, these computers aren’t very good at understanding this language because there isn’t much data available. The authors created a big dataset with lots of different types of Nepali speech and used it to train computer models to improve their accuracy. They also made the models work better by adding more variety to the training data. This helped the models reduce mistakes when transcribing speech, making them more accurate. This research shows how important it is to have high-quality data and a diverse range of examples for computers to learn from. |
Keywords
» Artificial intelligence » Data augmentation » Fine tuning
