Specialized Sound Recognition for Edge Devices
A lightweight sound recognition engine designed for smart cameras and IoT devices. Models as small as 0.2–1 MB, runs offline, data never leaves the device.
Supported Sound Types
Nine production-ready sound recognition engines for a wide range of detection scenarios
Baby Cry Detection
High-accuracy infant cry recognition for baby monitors, in-car child presence detection, and more
Snore Detection
Precise snore event detection for sleep monitors, smart mattresses, and health wearables
Dog Bark Detection
Reliable dog bark detection for smart doorbells, security cameras, and pet monitors
Cat Sound Detection
Detect cat meows and vocalizations for pet care devices and smart cameras
Glass Break Detection
Identify glass breaking sounds for security systems and retail anti-theft
Alarm Sound Detection
Detect smoke alarms, car alarms, and other alert sounds for security monitoring
Gunshot Detection
Real-time gunshot detection for smart city, public safety, and security systems
Knock Detection
Detect door knocks and tapping sounds for smart doorbells and home security
Scream Detection
Detect human screams and shouts for personal safety and security monitoring
Why soundSDK
Ultra Lightweight
INT8 models from 0.2–1 MB, CPU usage under 50MHz. Runs on resource-constrained embedded devices without GPU or NPU; inference latency and model tier adapt to your platform.
Data Privacy
All inference runs locally, no internet required. GDPR-compliant by design, ideal for home monitoring scenarios.
Cross-Platform
Supports ARM Linux, MIPS, x86 and more. Pre-adapted for HiSilicon, Ingenic, and other mainstream camera SoCs.
Quick Integration
Standard C API, run demo in 5 minutes. Comprehensive documentation and sample code reduce integration overhead.
Performance Benchmarks
| Sound Type | Accuracy | False Alarm | Model Size |
|---|---|---|---|
| Baby Cry | 97.2% | <2% | 0.2–1 MB |
| Snore | 96.5% | <2% | 0.2–1 MB |
| Dog Bark | 96.0% | <2% | 0.2–1 MB |
| Cat | 95.5% | <2% | 0.2–1 MB |
| Glass Break | 96.5% | <2% | 0.2–1 MB |
| Alarm | 96.0% | <2% | 0.2–1 MB |
| Gunshot | 95.8% | <2% | 0.2–1 MB |
| Knock | 95.0% | <2% | 0.2–1 MB |
| Scream | 95.5% | <2% | 0.2–1 MB |
* Test conditions: 16kHz sample rate, -10dB to +10dB SNR noise environment. Model size covers 0.2–1 MB INT8 variants; inference latency and model tier are configurable to match platform resources — stronger platforms can run larger variants with lower latency.
How It Works
Try Online
Upload audio files and experience the detection quality
Get SDK
Receive a 30-day free trial with full SDK and technical support
Integrate & Test
Integrate the SDK on your target device and validate performance
License
Per-device licensing with volume pricing for long-term partnership
Ready to add sound intelligence to your device?
30-day free trial, full-featured SDK, professional technical support.
Apply for Free Trial