Pretrained speech models detect wake words without extra training
Toward Elastic Speech Inference: Training-Free Wake-Word Detection from Pretrained ASR
Sound
Summary
Wake-word detection lets devices recognize a specific word to start listening. Usually, speech models need extra training to detect these wake words well. The authors show it’s possible to use existing large speech recognition models to detect wake words without additional training. They also find that these models can be made smaller without losing much accuracy, making them more efficient to run.
What this means in practice
- •For voice assistant developers: Implement wake-word detection using pretrained ASR models without extra training, simplifying deployment in varied environments.
- •For mobile device engineers: Reduce the size of speech recognition components for wake-word detection by half, enabling efficient use in resource-constrained devices.
Authors
Hwayeon Kim, Youngwon Choi, Hyeonyu Kim
Abstract
Recent ASR development has placed growing emphasis on generalization across diverse domains and acoustic conditions. Existing approaches typically adapt pretrained ASR models to front-end functions such as wake-up word (WuW) detection through additional training or task-specific modules. In this work, we explore the use of a shared pretrained ASR backbone for WuW detection without gradient-based fine-tuning and examine whether a compact encoder can be extracted using the PCA-based structured pruning approach of SliceGPT. Experiments with Parakeet-TDT-0.6B-v3 and Moonshine-base show that WuW detection performance remains relatively stable when the encoder channel dimension is reduced by 50%. These results suggest that task-relevant compact encoders can be derived from pretrained ASR models without fine-tuning.