Weakly Supervised Data Refinement and Flexible Sequence Compression for Efficient Thai LLM-based ASR

2025-05-28

AI Summary

To address the dual challenges of scarce high-quality labeled data and high computational overhead in low-resource Thai automatic speech recognition (ASR), this paper proposes EThai-ASR—the first large language model (LLM)-driven efficient Thai ASR system. Methodologically, we design a self-evolving weak-label refinement strategy to enhance speech encoder robustness; introduce a plug-and-play trimodal sequence compression module that enables dynamic-length compression, significantly reducing computation while preserving modeling capacity; and construct an end-to-end architecture comprising a speech encoder, a connector module, and a Thai-specific LLM-based decoder. Evaluated on multiple Thai ASR benchmarks, EThai-ASR achieves state-of-the-art performance. Furthermore, we publicly release a high-quality refined transcription dataset, establishing a new paradigm and foundational resource for low-resource speech recognition.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Despite remarkable achievements, automatic speech recognition (ASR) in low-resource scenarios still faces two challenges: high-quality data scarcity and high computational demands. This paper proposes EThai-ASR, the first to apply large language models (LLMs) to Thai ASR and create an efficient LLM-based ASR system. EThai-ASR comprises a speech encoder, a connection module and a Thai LLM decoder. To address the data scarcity and obtain a powerful speech encoder, EThai-ASR introduces a self-evolving data refinement strategy to refine weak labels, yielding an enhanced speech encoder. Moreover, we propose a pluggable sequence compression module used in the connection module with three modes designed to reduce the sequence length, thus decreasing computational demands while maintaining decent performance. Extensive experiments demonstrate that EThai-ASR has achieved state-of-the-art accuracy in multiple datasets. We release our refined text transcripts to promote further research.

Problem

• Addresses high-quality data scarcity in Thai ASR
• Reduces computational demands in LLM-based ASR
• Enhances weak labels via self-evolving refinement

Innovation

• Self-evolving data refinement for weak labels
• Pluggable sequence compression module
• LLM-based Thai ASR system