A Consortium Blockchain Framework for Low-Storage Coal Mine Dispatch Speech System

Chao Cheng, Huilin Wang, Wei Chen, Jueting Liu, Zehua Wang, Xin Li

Abstract: In coal mine communication systems, dispatch speech data contains critical production and safety information, yet its large volume and high confidentiality pose significant challenges for secure storage and efficient supervision. Traditional centralized storage methods suffer from risks of data tampering, opaque regulatory processes, and excessive storage costs. This study proposes a framework integrating LoRA-optimized Whisper ASR and GPT-4 based text summarization with a consortium blockchain for efficient speech management. The fine-tuned Whisper model, reduces CER by 59.6% in mining scenarios. GPT-4 generates concise summaries using multi-prompt strategies, extracting critical operational and environmental parameters. Summaries are securely stored in a consortium blockchain via private data collections, with hashes ensuring integrity. Improve regulatory transparency by decentralizing audits and tamper-proof records to address safety and efficiency gaps in mine data management. Experiments show that compared with the original text, the length of the summary text is reduced by 90%. Storing summaries (instead of speech text) can reduce CPU and memory overhead by 15.1% and 8%, respectively, and reduce communication traffic by 34%, while storage consumption is 99.7% lower than that of raw speech.

External IDs:dblp:conf/icic/ChengWCLWL25