Speech privacy can be tracked online by comparing different speech recognizers
CLEAR: Online Speech Content Leakage Estimation through Cross-ASR Disagreement
Sound
Summary
Speech privacy tools try to hide what people say while keeping other sounds intact. Usually, these tools are set up once and don’t change even if the chance of someone understanding the speech changes. The authors show a way to estimate how much speech is still understandable by comparing differences between several speech-to-text programs. When these programs disagree more, it means more privacy. This method works without needing the correct transcript and can help adjust speech privacy in real time.
What this means in practice
- •For mobile device developers: Enable real-time adjustment of speech privacy settings on phones by estimating how much speech content can be recovered without needing true transcripts.$Commercial implications: This paper enables privacy-aware voice features on consumer devices by dynamically estimating speech exposure during use, improving user privacy controls.
- •For smart home system integrators: Provide feedback on residual speech exposure from privacy filters in smart speakers to balance speech privacy with correct operation of voice commands.
Authors
Bhawana Chhaglani, Tanvi Kandepuneni, Jeremy Gummeson, Prashant Shenoy
Abstract
Signal-level speech privacy mechanisms suppress linguistic content while preserving acoustic information needed by downstream sensing applications. However, their privacy settings are typically evaluated/selected offline and remain fixed during deployment, even though speech-content leakage can vary substantially across utterances and speakers. Adapting privacy protection at runtime requires estimating how much speech remains recoverable, but conventional measures such as WER or PER require ground-truth transcripts and therefore cannot be computed online. We present CLEAR, a reference-free approach for estimating speech-content leakage at runtime using disagreement among heterogeneous ASR systems. Our key insight is that independently trained ASRs exhibit consistent behavior when linguistic content remains recoverable and increasingly disagree as privacy transformations obscure speech. Using configurable speech-suppression mechanism, we show that cross-ASR disagreement closely tracks transcript-grounded leakage across privacy operating points, achieving a correlation of 0.8. We further characterize the latency-accuracy trade-off of heterogeneous ASR subsets and use their hypotheses to identify potentially exposed words. These capabilities enable privacy to be treated as a runtime property rather than a fixed configuration: CLEAR can communicate residual speech exposure to users and provide feedback for dynamically adjusting privacy aggressiveness while retaining acoustic utility for downstream sensing tasks.