~/talk

Audio Keyword Reconstruction from On-Device Motion Sensor Signals via Neural Frequency Unfolding

on YouTube 06m
Talkat UbiComp 2021 · 2021-09-21..09-26· [security, machine-learning]

speakers

name:Tianshi Wang
links:none on file
name:Shuochao Yao
links:none on file
name:Shengzhong Liu
links:none on file
name:Jinyang Li
links:none on file
name:Dongxin Liu
links:none on file
name:Huajie Shao
links:none on file
name:Ruijie Wang
links:none on file
name:Tarek Abdelzaher
links:none on file

resources

video:https://youtu.be/UTiHEC_yk0o
slides:none on file

summary

Claims that a neural network reconstructs high-frequency spoken keywords from 100–400 Hz accelerometer and gyroscope traces by reversing aliasing in the time–frequency domain. It builds a frequency-expanding model using multi-resolution STFT, frequency convolutions and an alias unfolding layer to 16× upsampling of motion data, tested on a Google Pixel 1. Results beat three baselines on two datasets for pre-trained keywords, demonstrating feasible leakage but still limited generalization.

AI-written summary · about

related talks

~/search
type to search 6,249 conferences, 138,987 speakers and 185,172 talks
~/help
cmdctrl k
search every conference, speaker and series
/
focus the filter search
h
home
c
conferences
s
speakers
e
series
t
light / dark — the status bar button does the same
?
this help
esc
close this, then clear every filter — or just the search, while you are typing in it

Ignored while you are typing. Pages without a filter rail have no search to focus.