Sound Type Definitions

Currently Supported Sound Types

The SDK currently supports nine sound detection models, each as a standalone module:

Baby Cry Detection

Detects infant crying in complex home environments, distinguishing it from TV noise, conversations, and household appliances. Ideal for baby monitors, smart cameras, and in-car child presence detection (CPD).

• Generalized across different infant ages (newborn to toddler)

• Robust in -10dB to +10dB SNR noise conditions

• False alarm rate < 2%

• Test set: 12,000+ audio samples

Snore Detection

Detects snoring events, distinguishing them from normal breathing and body movement sounds. Suitable for sleep monitors, smart mattresses, and health wearables.

• Generalized across different ages and genders

• Robust against continuous background noise (AC, fans)

• Outputs onset/offset timestamps per snore event

• False alarm rate < 2%

• Test set: 8,000+ audio samples

Dog Bark Detection

Detects dog barking, distinguishing it from other animal sounds (bird calls, cat sounds) and environmental noise. Ideal for smart doorbells and outdoor security cameras.

• Works across different dog breeds

• Maintains accuracy in outdoor wind and rain conditions

• Supports 5-10 meter detection range

• False alarm rate < 2%

• Test set: 10,000+ audio samples

Additional Engines

Cat Sound Detection, Glass Break Detection, Alarm Sound Detection, Gunshot Detection, Knock Detection and Scream Detection follow the same module structure — the same API, the same integration flow, the same deployment guide. See each product page for scenarios and performance details.

Model Architecture

All models share a common architecture:

1. Frontend: Mel spectrogram (225×225, 64 Mel bins, 16kHz)

2. Backbone: Lightweight CNN with depthwise separable convolutions

3. Output: 2-class softmax (target sound / not target sound)

Models are trained with data augmentation (noise injection, time stretching, pitch shifting) for robustness. See our benchmark blog post for detailed evaluation results.

Multi-Type Detection

To detect multiple sound types simultaneously, run multiple SDK instances or configure a single instance with the appropriate model. Contact us for custom multi-type model development.