Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

Kevin Wilkinghoff, Alessia Cornaggia-Urrigshardt, Zheng-Hua Tan

TL;DR

This work tackles robust open-set few-shot keyword spotting in noisy environments by addressing threshold estimation, a key bottleneck for detection performance. It introduces a two-step score calibration that leverages embedding quantization and quantization-error normalization, combined as a single calibration gamma, to stabilize DTW-based keyword detections. Evaluations on kws-DailyTalk with hf radio channel simulations show that calibration improves test performance in noise and reduces sensitivity to threshold selection, while preserving validation performance. The findings indicate that both steps are complementary and essential for robust, noise-tolerant KWS in resource-constrained, real-world scenarios.

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.

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

TL;DR

This work tackles robust open-set few-shot keyword spotting in noisy environments by addressing threshold estimation, a key bottleneck for detection performance. It introduces a two-step score calibration that leverages embedding quantization and quantization-error normalization, combined as a single calibration gamma, to stabilize DTW-based keyword detections. Evaluations on kws-DailyTalk with hf radio channel simulations show that calibration improves test performance in noise and reduces sensitivity to threshold selection, while preserving validation performance. The findings indicate that both steps are complementary and essential for robust, noise-tolerant KWS in resource-constrained, real-world scenarios.

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.
Paper Structure (14 sections, 5 equations, 3 figures, 2 tables)

This paper contains 14 sections, 5 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Template-based KWS with DTW
  3. Quantization-based score calibration
  4. Step 1: Template quantization
  5. Step 2: Quantization error-based score normalization
  6. Combined score calibration approach
  7. Experimental evaluation
  8. Dataset
  9. Implementation details
  10. Effectiveness of the proposed score calibration approach
  11. Quality assessment of estimated decision thresholds
  12. Effect of individual steps
  13. Limitations and future work
  14. Conclusion

Figures (3)

  • Figure 1: Illustration of combining the embeddings belonging to different segments of a single recording. Adapted from wilkinghoff2024audio.
  • Figure 2: Illustration of the quantization steps. Without quantization, the embeddings are directly compared (left). With quantization, the closest centers are compared (middle). When normalizing the scores, the distance between the embeddings is adjusted relative to their quantization errors (right).
  • Figure 3: Difference between the optimal and estimated thresholds (left) and between the performances obtained with these thresholds (right) on the test set of kws-DailyTalk for different snr when calibrating and not calibrating the scores. $95\%$ confidence intervals over five independent trials are shown.