Article Details

Abstract

Aiming at the mandatory requirements of standardized operation instruction interaction and post-event safety analysis and processing in China's railway transportation safety management rules, a technical scheme of Automatic Speech Recognition (ASR) of railway operation speech based on Transformer is proposed, and the ASR system of railway operation is realized. The system makes full use of the multi-head self-attention mechanism of the Transformer architecture, which can effectively capture long-distance dependencies and facilitate parallel computing. According to the characteristics of the standard language of railway operation specifications, the training is optimized, which significantly improves the performance of ASR and improves the recognition accuracy of the system to more than 90%. The system realizes the automatic text transfer function of safety analysis after dialogue in railway operation process, and meets the practical application requirements of safety supervision in railway department.

Keywords

railway operation voice; ASR; Transformer model

References

1. [1] National Railway Administration Promote the high-quality development of railways and contribute to the economic growth of the "railway" force [EB/OL]. (2024-12-26) [March 3, 2025] https://www.nra.gov.cn/xwzx/xwxx/xwlb/202412/t20241226_347601.shtml.
2. [2] Dayang Network. In 2024, China's railways transported 4.08 billion passengers and 3113 kilometers of new lines were put into operation. [EB/OL]. (2025-01-02) [2025-03-01] https://news.dayoo.com/gzrbrmt/202501/02/170628_54769704.htm.
3. [3] Li Shiwang, Jiang Lin, Wang Guilin Research on Intelligent Dialogue System for Railway Passenger Transport Marketing Analysis Based on Natural Language Processing [J]. Railway Computer Applications, 2024, 33 (8): 61-71
4. [4] Chen Nan, Cao Xuehong, Jiao Liangbao, etc Text Error Correction Algorithm for Power Inspection Voice Instruction Recognition [J]. Computer and Digital Engineering, 2022, 50 (1): 116-123+134
5. [5] Mao Hongliang Research on Text Error Correction Method for Speech Recognition in the Field of Power Grid Dispatching [D]. Urumqi: Xinjiang Normal University, 2023
6. [6] Ma Wenhui, Feng Guobin, Liu Weimin, etc Research on Text Error Correction Algorithm after Speech Recognition [J]. Railway Communication Signal, 2020, 56 (11): 55-58
7. [7] Ma Xiaoliang, An Lingling, Deng Congjian, etc Optimization technology for automatic speech transcription based on industry vocabulary [J]. Journal of South China University of Technology: Natural Science Edition, 2023, 51 (8): 118-125
8. [8] ZHANG Q，LU H，SAK H，et al. Transformer Transducer: A Streamable Speech Recognition Model with Transformer Encoders and RNN-T Loss [C]//ICASSP 2020 - 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP).Barcelona：IEEE，2020：7829-7833.
9. [9] GULATI A，QIN J，CHIU C C，et al. Conformer: Convolution-augmented Transformer for Speech Recognition [J/OL].arXiv:2005.08100 [eess.AS].[2025-01-05].https://doi.org/10.48550/arXiv.2005.08100.
10. [10] XU M L，LI S Q，ZHANG X L. Transformer-based end-to-end speech Recognition with Local Dense Synthesizer Attention [C]//ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP).Toronto：IEEE，2021：5899-5903.
11. [11] LI M H，ZHANG S C，ZORILĂ C，et al. Transformer-Based Streaming ASR with Cumulative Attention [C]//ICASSP 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP).Singapore：IEEE，2022：8272-8276.
12. [12] VASWANI A，SHAZEER N，PARMAR N，et al. Attention is All You Need [J].Proceedings of the 31st International Conference on Neural Information Processing Systems.Red Hook：Curran Associates Inc.，2017：6000-6010.

About the authors

Zhu Jianrong (1978.03-), male, Han nationality, from Qingyuan, Guangdong, engineer, bachelor's degree, research direction: artificial intelligence applications.