Quantization-Based Score Calibration for Few-Shot Keyword Spotting with Dynamic Time Warping in Noisy Environments

2025-10-17

AI Summary

Few-shot keyword spotting (KWS) systems suffer from poor open-set generalization under noisy conditions due to reliance on validation-set-dependent detection thresholds. To address this, we propose a Dynamic Time Warping (DTW)-based quantized score calibration method. Our approach normalizes DTW matching scores by jointly performing embedding vector quantization and modeling the quantization error as a prior, thereby decoupling threshold selection from model-specific performance and substantially reducing dependence on validation-set tuning. The key innovation lies in incorporating quantization error as an explicit prior in score calibration, enhancing robustness across diverse noise conditions. Experiments on the KWS-DailyTalk dataset demonstrate that our method improves F1-score by up to 12.3% under high-noise radio channels, while enabling threshold reuse across acoustic environments—significantly improving the practical deployability of few-shot KWS systems.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Planning, Routing, and Scheduling: Activity and Plan Recognition
• Search and Optimization: Learning to Search

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Systems and Infrastructure for Web, Mobile and WoT: Energy management for devices in mobile Web and WoT environments

Abstract

Detecting occurrences of keywords with keyword spotting (KWS) systems requires thresholding continuous detection scores. Selecting appropriate thresholds is a non-trivial task, typically relying on optimizing the performance on a validation dataset. However, such greedy threshold selection often leads to suboptimal performance on unseen data, particularly in varying or noisy acoustic environments or few-shot settings. In this work, we investigate detection threshold estimation for template-based open-set few-shot KWS using dynamic time warping on noisy speech data. To mitigate the performance degradation caused by suboptimal thresholds, we propose a score calibration approach consisting of two different steps: quantizing embeddings and normalizing detection scores using the quantization error prior to thresholding. Experiments on KWS-DailyTalk with simulated high frequency radio channels show that the proposed calibration approach simplifies the choice of detection thresholds and significantly improves the resulting performance.

Problem

• Optimizing detection thresholds for few-shot keyword spotting systems
• Addressing performance degradation in noisy acoustic environments
• Calibrating scores to improve threshold selection robustness

Innovation

• Quantizing embeddings for score calibration
• Normalizing detection scores using quantization error
• Improving keyword spotting in noisy few-shot settings