Adaptive layer reduces false wake ups in voice assistants
Not All Speech Is Intent: Adaptive Self-Correcting Inference Layer for Post-ASR False Wake-Up
Computation and LanguageArtificial Intelligence
Summary
Sometimes voice assistants mistakenly think they’ve been called when a person actually didn’t mean to speak to them, causing annoying errors. The authors developed a new system called ASCIL that listens again after the assistant is triggered and uses clues like hesitation or silence to check if the wake-up was intentional. It learns from past mistakes and adjusts itself over time without needing people to label the errors manually. Their tests showed ASCIL decreased these false alarms by over half while keeping the assistant’s responses quick and accurate.
What this means in practice
- •For voice assistant developers: Improve wake word detection systems by integrating a feedback-driven layer that reduces accidental activations from phonetically similar speech.
- •For smart home device manufacturers: Enhance user satisfaction by decreasing false wake-ups in voice-controlled devices using adaptive post-ASR correction without adding noticeable delay.$Commercial implications: Enables smarter voice control products with fewer errors, appealing directly to consumers seeking reliable smart home interactions.
Authors
Preeti Saraswat, Divya Neelagiri, Anil Yadav
Abstract
False wake-up activations remain a persistent challenge in conversational AI. Speech phonetically similar to a device's wake word can produce a syntactically valid and semantically coherent ASR transcript that the assistant incorrectly executes. Most existing systems make a single intent decision in isolation, without a mechanism to learn from recurring errors over time or adapt to individual users through personalized learning. We introduce the Feedback-Driven Adaptive Self-Correcting Inference Layer (ASCIL), a complementary post-ASR correction framework that re-evaluates wake-up intent before response generation by fusing acoustic embeddings, linguistic cues, device context, and patterns from past misclassifications. ASCIL interprets implicit signals, including hesitation, disengagement, and silence, and explicit signals, including cancellation and repetition, as automatically inferred, noisy behavioral indicators of potential misclassification. These signals drive online pattern updates without manual annotation, whereas the intentional/unintentional reference labels used for offline evaluation are human-annotated. It generalizes from prior errors, applies corrective adjustments at inference time, and continuously updates in parallel with natural-language execution. Evaluated on a proprietary dataset of 3,667 interactions with human-annotated intentional/unintentional reference labels spanning 14 acoustic and contextual conditions, ASCIL achieves 54.27% relative error reduction on a session-disjoint subset constructed from baseline failures, and up to 24.39% relative error reduction at threshold 0.90 on the issue-tagged evaluation slice. These gains are achieved while improving intentional acceptance rates, with a median added latency below 60 ms in the reported benchmark.