Calibrated order-aware method improves open-vocabulary keyword spotting accuracy
CORD-KWS: Calibrated, Order-Aware Detection for Open-Vocabulary Keyword Spotting
Sound
Summary
Detecting spoken keywords without retraining for each new word is hard, especially in difficult audio conditions. The authors found that existing models often fail because they do not calibrate their confidence scores properly, making it hard to apply a fixed detection threshold. They propose a new training method, called CORD-KWS, that calibrates these scores and also helps the model learn the correct order of sounds in words. This approach improves detection accuracy on challenging test sets without increasing the time it takes to recognize keywords.
What this means in practice
- •For voice interface developers: Improve voice-controlled assistants to detect new keywords accurately without retraining or heavy computation.$Commercial implications: Enables real-time keyword detection products that adapt to diverse user commands without costly model updates.
- •For speech recognition engineers: Deploy keyword spotting in low-resource or embedded systems with limited computational power by using efficient calibrated models.
Authors
Ramesh Gundluru, Adarsh Arigala, Sri Rama Murty Kodukula
Abstract
Open-vocabulary keyword spotting (KWS) must detect arbitrary keywords without retraining. Cross-attention-based models achieve state-of-the-art performance but require pairwise interaction between the audio and each keyword, recomputing the fused representation for every audio--keyword pair. Embedding-based models avoid this computational cost through independent encoding and similarity scoring, yet remain inferior on challenging benchmarks such as LibriPhrase-hard. We show that this gap is primarily a training issue rather than a limitation of model capacity. While contrastive learning encourages correct keywords to rank above negatives, it does not explicitly calibrate absolute similarity scores, which are critical for applying a fixed threshold across keyword detection. We propose Calibrated, Order-aware Detection KWS (CORD-KWS), a training framework that introduces two complementary objectives while retaining the same encoders, cosine scoring, and O(1) keyword enrollment: a calibrated detection head that supervises absolute scores, and a frame-level CTC objective that preserves phonetic order information. CORD-KWS achieves EERs of 0.43% and 9.64% on LibriPhrase-easy and LibriPhrase-hard, respectively, outperforming existing embedding- and cross-attention-based models.