Embedded Files
Introduction
Introduction
This is the demo homepage of NormDetect. We present 3 aspects in the following: Demo, Samples, and Real-time Algorithm.
- Demo
We implemented a NormDetect APP, which is designed to distinguish the inaudible voice attacks. Moreover, we set it with 3 modes: Non-speech/Normal/Attack. (shown in the right side, more details are in Demo 1)
We also show 3 demo videos as below, they represent a) near-field detection (60cm), b) far-field detection (150cm), and c) far-field detection with different attacking angles, respectively. Three smartphones run the NormDetect APP simultaneously in each scenario, and we display normal voice and then inaudible voice attacks. NormDetect performs well in these scenarios.
PS: we recommend that you watch these demos on full screen. Thanks for your watching!
Identifying Near-field Attacks
Settings:
Near-field (60cm)
angle (0°)
Ambient Noise (55 dB)
Recording Devices: Google Nexus5
Identifying Far-field Attacks
Settings:
Far-field (150cm)
angle (0°)
Ambient Noise (55 dB)
Recording Devices: Google Nexus5
Identifying Attacks from different angles
Settings:
Far-field (150cm)
angle (30°)
Ambient Noise (55 dB)
Recording Devices: Google Nexus5
2. Samples
2. Samples
3. Real-time Algorithm
3. Real-time Algorithm
We generate a random set of audio sequences and place the attack audio of the same content in the front and the normal sound in the back. Normdetect calculates each audio with an anomaly score in real-time and plots it on the coordinates. We set the red asterisk for abnormal, the blue dot for normal, and the green horizontal line for normal and abnormal judgment threshold.
Page updated
Google Sites
Report abuse