Audio models

This page adds audio checks to the test stages. Refer to Ready Models and Ready Model Reports for the range, confusables, and test method of each model.

Signal chain checks

The sample rate, channels, and bit depth in your firmware match the model report.
The audio sounds good, with no audible distortions or glitches.
The noise floor is low during silence, with no audible hum from the hardware.
The loudest expected event stays below clipping. Clipping flattens the waveform peaks and removes features the model uses.
The microphone signal goes straight to the model, with no AGC, resampling, or filtering.

Factors that affect performance

Masking noise: background sound at similar frequencies can cover the target. Speech, television, and changing machinery noise have the most effect.
Reverberation: hard surfaces such as concrete, glass, and tile blur the timing of the target sound.
Clipping: very loud events flatten the waveform, and the model cannot recover the original shape.
Low SNR: the target sound is quiet compared with the background.

Measure SNR

The report states the minimum SNR the model needs. Record the SNR of each scenario so you can compare it with that minimum. Use one of these methods:

Sound level meter: measure the target level at the microphone position, then the background level. SNR (dB) ≈ target level − background level.
Recording: calculate the RMS of an event window and of a background window from the same recording. SNR (dB) = 20 × log₁₀(RMS event / RMS background).

Placement and environment

Test at three or more distances (near, mid, and far) and at least two angles (front and off-axis).
Test with the enclosure fitted. The enclosure changes the sound that reaches the microphone.
Test in at least two environments: a quiet baseline and the intended deployment environment.
Include at least one changing background, such as television, conversation, or machinery.

Audio test matrix

Copy this table and adapt it to your test. The middle part of each ID is the stage. Priority is Core (recommended for every test plan) or Optional (add where it fits your product). Rows with a pass/fail result follow Sample size. Rows that record how the model behaves do not need a fixed number of test events.

ID	Priority	Purpose	Target	Placement	Background	Expected result
A-1-01	Core	Noise floor check	None	Intended mount	Silence, 5 to 10 s	Low noise floor, no audible hum
A-1-02	Core	Signal check	Target sound	0.3 to 0.5 m	Silence	Clear signal, no clipping
A-2a-01	Core	Baseline detection	Target event	Close range	Silence	Detects the target across the planned test events
A-2a-02	Core	Baseline false positives	None	Intended mount	Quiet, 5 min	No or very few false positives
A-2b-01	Core	Intended placement	Target event	Deployment distance and mount	Silence	Detection rate close to baseline
A-2b-02	Optional	Maximum range	Target event	Maximum range in the report	Silence	Compare with A-2b-01
A-2b-03	Optional	Enclosure effect	Target event	Intended distance	Silence	Compare with and without the enclosure
A-2c-01	Core	Confusables	None	Intended mount	Confusables listed in the report	False positives within your criteria
A-2c-02	Core	Changing background	None	Intended mount	Speech or TV, recommended false positive duration	False positives within your criteria
A-3a-01	Core	Deployment environment	Target event	Deployment conditions	Typical background	Meets your acceptance criteria
A-3a-02	Core	Deployment false positives	None	Deployment conditions	Typical background, recommended false positive duration	False positives within your criteria
A-3b-01	Core	Range limit	Target event	Near maximum range	Typical background	Record the detection rate
A-3b-02	Optional	Noisiest environment	Target event	Intended mount	Noisiest realistic environment	Record the detection rate
A-3b-03	Optional	Muffled target	Target event, muffled, for example into a blanket or pillow	Intended mount	Typical background	Record how the model behaves
A-3b-04	Optional	Device handling	None	Device rubbed, tapped, or dropped	Typical background	Record how the model behaves