Effective Integration of KAN for Keyword Spotting

2024-09-13

arXiv.org

AI Summary

To address the performance limitations of keyword spotting (KWS) on resource-constrained smart devices, this paper introduces the Kolmogorov–Arnold Network (KAN) to speech-based KWS for the first time, proposing a synergistic modeling framework that integrates KAN with a lightweight 1D CNN. The method leverages KAN’s superior expressivity for low-dimensional, high-level semantic features while exploiting CNN’s efficiency in capturing local time-frequency patterns. We design multiple learnable ensemble strategies and perform end-to-end training using standard acoustic features—e.g., MFCCs and log-Mel spectrograms. Evaluated on mainstream KWS benchmarks (e.g., Google Speech Commands), the proposed model significantly outperforms pure CNN baselines, achieving absolute accuracy gains of 2.3–4.1% and demonstrating enhanced robustness. This work validates KAN’s effectiveness and potential for lightweight, temporal speech modeling in edge-deployable KWS systems.

Technology Category

• Natural Language Processing: Speech
• Machine Learning: Kernel Methods
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs

Abstract

Keyword spotting (KWS) is an important speech processing component for smart devices with voice assistance capability. In this paper, we investigate if Kolmogorov-Arnold Networks (KAN) can be used to enhance the performance of KWS. We explore various approaches to integrate KAN for a model architecture based on 1D Convolutional Neural Networks (CNN). We find that KAN is effective at modeling high-level features in lower-dimensional spaces, resulting in improved KWS performance when integrated appropriately. The findings shed light on understanding KAN for speech processing tasks and on other modalities for future researchers.

Problem

• Keyword Recognition
• Smart Devices
• Speech Assistants Performance

Innovation

• Kolmogorov-Arnold Network
• Keyword Spotting
• 1D Convolutional Neural Network