
Smart Home
Security & Monitoring
Security cameras
Video doorbells
Building surveillance
Home Entertainment
Smart speakers
Remote controls
Set-top boxes
Smart displays
Televisions
Video entertainment
Home Controls
Door locks
Thermostats
Smart lighting
Motion & pressure sensors
Smart Appliances
Robots
White goods
Fitness Equipment
Sample Applications Deployed in Complex Environments
Speech as an interface
Noise suppression
Echo cancellation
Glass break detection
Face recognition
Person & package detection
Vehicle detection
Autoframing
Gesture recognition
Pose classification
Motion & pressure sensors
T3/T4 alarms
Low-battery alerts
Use Case: Robust Window Glass Break Detection
Deep learning enables the detection of external window breaking while ignoring other noises within the environment of the device.
-
Existing solutions look only for a noise level at a specific frequency range.
-
This misses many glass break events (false negatives). and triggers other loud noises that are not glass (false positive).
Syntiant has collected thousands of examples of external home glass breaking and trained advanced neural networks to trigger only when an external glass break is detected while ignoring all other noises, including similar sounding events such as plates breaking in the kitchen.
Use Case: Pet Detection
Using NDP200 and a low-cost Pixart camera, trained up a MobileNetV1 0.25 person detection model.
-
Input size 320x240 grayscale.
-
Used pet/non-pet images from MS COCO for demo network.
-
Achieved 90% accuracy on the demo model after less than 5 hours of training.
**Pet detection on NDP200 can run for weeks on a single car battery.
Use Case: Local voice command of iRobot Roomba
On-device command of medium-sized home appliance, enabled by the NDP101®.
- Use of microphones with voice activity detection allows NDP to stay at ultra-low-energy standby mode and only wakes up when alerted by sound
Use Case: Hands Free Remote Control
The NDP120 adds a robust voice interface to a remote control while not draining the battery.
-
Noise Suppression and energy management: Low energy implementation of always-on wake word detection in a noisy home environment
-
Syntiant’s processors and models are based on advanced convolutional neural networks to trigger only when a wake word is heard
-
Noise suppression improves wake-word performance even at low SNR (signal to noise ratio) environments
-
Use of microphones with voice activity detection allows NDP to stay at ultra-low-energy standby mode and only wakes up when alerted by sound
-
*Wake word detection on NDP120 runs for 12 mo. on a 2x AA battery
Use Case: Speech as an Interface
Ideal for smart speakers: low power, hands-free operation, even in the presence of noise.
-
Far-field always-on voice allows for wake-word detection, even in the presence of very loud music at very low power: ideal for smart speakers.