Edge AI
Audio Edge Signals in Practice
Feature extraction, on-device VAD, and privacy-preserving buffers for voice-adjacent products.
Format: Weekend intensive · Level: Foundational · Duration: 3 weeks
Tuition (informational): 610,000 KRW
Lead mentor
Shipped on-device perception for warehouse robotics; cares about honest failure modes.
Overview
Voice-adjacent products need discipline around buffers and consent cues. You implement voice activity detection that is explainable, audit buffers with TTLs, and document deletion paths. Labs avoid raw voice storage; synthetic chirps stand in for identifiable audio.
What is included
- VAD thresholding with plotted false positives
- Ring buffer TTL policy workshop
- On-device feature extraction without cloud round-trips
- LED/buzzer consent cue patterns
- Privacy review checklist for Korean deployments
- Latency budget spreadsheet
- Dry-run incident for buffer leak
Outcomes
- Ship a VAD configuration sheet with measured false accept rate
- Demonstrate buffer TTL enforcement
- Write a deletion path note for support teams
Participant notes
-
Audio Edge Signals in Practice forced us to document TTLs before marketing promised “always listening.” Consent cue patterns are shipping.
Camille D. · Product prototyper · BlueRiver Group · 5/5