This is a page that curates AI-related papers published worldwide. All content here is summarized using Google Gemini and operated on a non-profit basis. Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.
ECHO: Frequency-aware Hierarchical Encoding for Variable-length Signals
Created by
Haebom
Author
Yucong Zhang, Juan Liu, Ming Li
Outline
This paper proposes a novel foundational model, ECHO, that focuses on general machine signal modeling for various machine signals (e.g., acoustics, vibration, industrial sensor data) with arbitrary sampling rates. ECHO integrates an advanced band-splitting architecture and frequency-localized embedding to capture spectral localization under arbitrary sampling configurations. Furthermore, it integrates sliding patches to support variable-length inputs without padding or truncation, generating concise embeddings that preserve temporal and spectral fidelity, making it naturally extendable to streaming scenarios. Experimental results on various machine signal datasets, including the DCASE challenge (2020-2025) and widely used industrial signal corpora, demonstrate consistent state-of-the-art performance in machine signal anomaly detection and fault classification, validating the effectiveness and generalization ability of the proposed model. ECHO was released in https://github.com/yucongzh/ECHO .
We present a general fundamental model for various machine signals with arbitrary sampling rates.
◦
It enables spectral localization and efficient processing through band-splitting architecture and frequency location embedding.
◦
Provides variable-length input processing and streaming support using sliding patches.
◦
Achieve state-of-the-art performance in machine signal anomaly detection and fault classification.
◦
It has been released as open source to increase accessibility.
•
Limitations:
◦
There is no mention of specific Limitations or future research directions in the paper.
◦
Although performance evaluations on various datasets have been presented, there may be a lack of analysis on the potential for performance degradation on specific datasets or limitations in generalization performance.
◦
A detailed analysis of the model's computational complexity and memory usage may be required.
