Noise

NOISE (dk.cachet.carp.noise) samples the sound level from the phone's microphone in short windows and adds one Noise measurement per window, with four statistics in decibels. It is an event-based measure on the phone, on Android and iOS.

The measure stores no sound: the probe reduces the microphone samples to levels as they arrive. The constant is on MediaSamplingPackage, and the levels come from the noise_meter plugin.

Add the measure to a protocol

NOISE schedules its own sampling windows, so start it once with an ImmediateTrigger(). This task samples for 5 seconds every 30 seconds instead of the default:

protocol.addTaskControl(
  ImmediateTrigger(),
  BackgroundTask(
    measures: [
      Measure(
        type: MediaSamplingPackage.NOISE,
        samplingConfiguration: PeriodicSamplingConfiguration( 
          interval: const Duration(seconds: 30),
          duration: const Duration(seconds: 5),
        ),
      ),
    ],
  ),
  phone,
);

A shorter interval gives more measurements, but the microphone is on more often and the battery drains faster.

Configuration

NOISE uses a PeriodicSamplingConfiguration:

Both properties are required when you create the configuration. The defaults in the table come from the package's default sampling scheme, which samples for 10 seconds every 5 minutes.

Sampling windows

While the task runs, the probe works in a cycle:

1. One interval after the task starts, and then once per interval, it opens a window. With the defaults, the first window opens 5 minutes after the task starts.
2. During the window, it listens to the microphone and turns each buffer of samples into a reading with two levels. On Android, a buffer holds 6,400 samples, about 145 milliseconds of sound at 44.1 kHz.
3. When the window closes, it computes the four statistics from the readings and adds one measurement. The measurement's sensorStartTime and sensorEndTime are the start and end of the window.

If the task stops during a window, the readings of that window are discarded. A window also adds no measurement when fewer than two of its readings have a finite level, because the standard deviation is then undefined. A window of digital silence has no finite readings, and the probe reports an error for it instead.

Decibel values

The microphone samples are scaled to values between -1 and 1. The two levels of a reading come from the lowest and the highest sample in its buffer:

mean level = 20 × log10(32768 × (|lowest sample| + |highest sample|) / 2)
peak level = 20 × log10(32768 × highest sample)

The values are decibels relative to the smallest step of a 16-bit sample, so full scale is about 90.3 dB. They are not calibrated sound pressure levels: the same sound gives different values on different phone models. Compare noise levels within the same phone, or calibrate each phone model against a sound level meter.

Platform differences

Limitations

• Not calibrated: the levels depend on the phone's microphone and differ between phone models.
• Background sampling on Android: Android 11 or later lets a background app use the microphone only from a foreground service of type microphone. The CAMS setup declares its foreground service as dataSync|specialUse, not microphone, so test background sampling on the phones of your study.
• Background sampling on iOS: sampling continues in the background only with the audio background mode.
• Delayed start: the first window starts one interval after the task starts, and a window in progress is discarded when the task stops.

Data

Each Noise holds the statistics of one window, computed from the readings with a finite level:

For the measurement envelope around the data, see Measurements and data. An example of the data:

{
  "__type": "dk.cachet.carp.noise",
  "meanDecibel": 47.83251874509615,
  "stdDecibel": 3.9171604375130887,
  "minDecibel": 41.06923398265123,
  "maxDecibel": 68.72041385620347
}