Evaluating and Improving Automatic Speech Recognition Systems for Korean Meteorological Experts

2024-10-24

arXiv.org

AI Summary

This work addresses the degraded automatic speech recognition (ASR) performance in Korean meteorological applications, caused by domain-specific terminology and linguistic complexity. To support Korean meteorological experts, we develop a specialized spoken-query ASR system. We introduce the first benchmark dataset for spoken queries in the Korean meteorological domain; propose a lightweight text-to-speech (TTS)-driven data augmentation method that improves recognition accuracy of meteorological terms without compromising general ASR capability; and conduct domain adaptation and evaluation based on multilingual Whisper ASR models. Experiments demonstrate a +12.3% absolute improvement in meteorological term recognition accuracy. Furthermore, we publicly release the dataset, evaluation framework, and augmentation methodology—establishing a reusable benchmark and practical technical pathway for vertical-domain ASR development in Korean.

Technology Category

• Natural Language Processing: Speech
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval
• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data

Abstract

This paper explores integrating Automatic Speech Recognition (ASR) into natural language query systems to improve weather forecasting efficiency for Korean meteorologists. We address challenges in developing ASR systems for the Korean weather domain, specifically specialized vocabulary and Korean linguistic intricacies. To tackle these issues, we constructed an evaluation dataset of spoken queries recorded by native Korean speakers. Using this dataset, we assessed various configurations of a multilingual ASR model family, identifying performance limitations related to domain-specific terminology. We then implemented a simple text-to-speech-based data augmentation method, which improved the recognition of specialized terms while maintaining general-domain performance. Our contributions include creating a domain-specific dataset, comprehensive ASR model evaluations, and an effective augmentation technique. We believe our work provides a foundation for future advancements in ASR for the Korean weather forecasting domain.

Problem

• Improving weather forecasting efficiency using ASR for Korean meteorologists.
• Addressing challenges in ASR for Korean weather domain-specific vocabulary.
• Developing and evaluating ASR models with domain-specific Korean datasets.

Innovation

• Integrated ASR for Korean weather queries
• Developed domain-specific Korean evaluation dataset
• Implemented text-to-speech data augmentation technique