Application of Hierarchical Architecture Technology in Edge Devices
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
With the deepening exploration of intelligent audio and the increasing demand for low latency, deploying deep learning models at the edge can not only effectively reduce latency, but also better protect user privacy. However, the high degree of hardware coupling of algorithms in traditional development models seriously restricts the portability and scalability of the system. With in-depth exploration of this issue, we propose a collaborative layered audio processing software architecture for multi-edge devices, which logically divides the various parts of the system into five relatively independent levels: hardware, drivers, data, algorithms, and applications. This eliminates the tight dependence of software and hardware and improves the flexibility and scalability of the system. At the same time, designing a heartbeat monitoring and asynchronous communication mechanism based on TCP/IP can achieve a stable and complete processing pipeline for VAD, streaming ASR, and punctuation recovery on a single edge node, which can also improve the reliability and synchronization efficiency of multi-device collaboration. In-depth experimental practice on Huawei Ascend Atlas 200I DK A2 has shown that it not only meets real-time performance requirements, but also has good system robustness and good scalability.
Fields of Study & Subjects
IoT and Edge/Fog Computing
Speech Recognition and Synthesis
Embedded Systems Design Techniques
Publication details
- Journal / source
- International Conference on Communication Networks and Machine Learning (CNML)
- Pages
- 1197–1200
- Publication date
- 30 January 2026
- Language
- EN