Squeezing keyword spotting into tiny models with circular time matching
CircleMatch: Prototype Matching with Circular Temporal Statistics for Tiny Keyword Spotting
Sound
Summary
Keyword spotting is the task of detecting specific spoken words, which is important for voice assistants and devices. The authors created CircleMatch, a very small model that can identify keywords using fewer than 7,000 parameters by compressing sound features and matching them against learned examples called prototypes. Their approach also uses a novel way to handle time information by encoding it as angles on a circle, which helps summarize when a word occurs. CircleMatch works well on English and Spanish datasets while keeping the model size extremely small.
What this means in practice
- •For mobile app developers: Implement compact keyword detection in apps to enable voice control using models with very small memory footprints and efficient temporal processing.
- •For wearable device engineers: Deploy extremely lightweight keyword spotters on resource-constrained wearables for offline voice command recognition with minimal power use.$Commercial implications: Enables voice interfaces on low-cost wearables by fitting keyword spotting into few parameters, making new products feasible.
Authors
Jiajun Sun, Zhe Gao
Abstract
Keyword spotting (KWS), the task of identifying predefined words in speech, is a core capability of voice-enabled devices. Achieving high KWS accuracy under tight parameter budgets across different vocabulary sizes remains challenging. We present CircleMatch, a matching framework enabling KWS with very few parameters. Its encoder independently compresses frequency bands and fuses them into frame features. These features are then matched against learned class-specific prototypes to produce temporal response curves. Parameter-free circular aggregation encodes time as angles and summarizes response distributions and relative timing for classification. We develop four tiny variants, Circle-D4, Circle-D8, Circle-D16, and Circle-D32, ranging from approximately 1k to 7k parameters in the 12-class setting. Experiments with multiple random seeds on Speech Commands v1/v2 and the English and Spanish Micro subsets of the Multilingual Spoken Words Corpus demonstrate competitive accuracy with tiny models. Our qualitative analysis further suggests approximate shift equivariance of prototype responses and adaptation to temporal compression. Code and model weights are available at https://github.com/ora942878/CircleMatch.