嵌入式关键词识别模型可扩展新增关键词，不丢旧词且计算量更低。
Scalable Keyword Spotting via Modular Network Expansion
- 冻结主干网络，仅训练轻量扩展分支添加新词。
- 新词误拒率降至4.37%，比基线降低2.09个百分点。
- 适合需持续更新关键词的边缘设备部署场景。
嵌入式设备上的关键词检测（KWS）模型在部署后常需新增关键词，但若原始训练数据不可用且不能容忍旧关键词性能下降，则更新极为困难。本文提出参数受限的模块化扩展方法：冻结基础网络（含批归一化统计与核心分类器），仅训练一个轻量级扩展分支及独立的新词头，保留原有关键词的输出逻辑、阈值和核心特征。在固定运行点下，新词平均误拒率从6.46降至4.37，优于参数匹配的独立模型基线；在≤10k新增参数预算下，乘加操作数（MACs）为16.34M，低于适配器（18.45M）与LoRA（20.52M）基线，实现高效更新与稳定性能。
原文摘要 · Abstract (English)
Keyword spotting (KWS) models on embedded devices often need to add new keywords after deployment, but updates are difficult when original training data are unavailable and regressions on existing triggers are unacceptable. At a fixed operating point, our method reduces average new-keyword false reject rate (FRR) from 6.46 to 4.37 versus a parameter-matched separate-model baseline and outperforms parameter-efficient tuning baselines (adapters, LoRA), while using fewer multiply-accumulate operations (MACs) under the same added-parameter budget ($\leq$10k): 16.34M vs 18.45M/20.52M. We achieve this via parameter-capped modular expansion: the base network, including batch-normalization statistics and the core classifier, is frozen, and only a lightweight expansion branch with a separate new-keyword head is trained, preserving core logits, shipped outputs, and thresholds for existing keywords.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。