How iOS 15 Sound Recognition Detects Custom Alarms

Apple introduced custom alarm and appliance detection in iOS 15 to expand accessibility, allowing users to train their iPhone to recognize unique environmental sounds. By capturing real-world audio samples directly through the device's microphone, iOS 15 utilizes on-device machine learning to generate a distinct acoustic profile. Once trained, the system continuously monitors ambient audio in the background and pushes an alert whenever a matching sound pattern is identified, providing essential awareness for individuals who are deaf, hard of hearing, or wearing headphones.

On-Device Audio Training

Custom sound detection begins with user-guided data collection. When setting up a custom alarm or appliance sound in the Accessibility settings, iOS prompts the user to record multiple iterations of the target noise (such as a microwave beep, washing machine chime, or unique smoke detector tone). Recording multiple instances allows the operating system to capture subtle variations in pitch, volume, duration, and background interference, creating a reliable baseline for comparison.

Spectrogram and Feature Extraction

During the recording phase, iOS 15 converts raw audio signals into spectrograms—visual representations of frequencies over time. The software breaks down the acoustic input into specific components:

• Frequency: The exact pitch range of the appliance or alarm tone.
• Duration: The length of individual beeps or tone bursts.
• Cadence and Rhythm: The repetitive pattern and intervals between sounds.
• Harmonic Structure: The distinct overtones that separate an electronic appliance chime from human voices or ambient background noise.

Machine Learning and Local Model Generation

Once the samples are captured, Apple's Neural Engine processes the data using Core ML frameworks entirely on the device. Instead of relying on a generalized preset, the operating system constructs a lightweight, customized machine learning model tailored specifically to that sound signature. This on-device processing ensures that no raw audio recordings are sent to external servers or the cloud, preserving user privacy.

Passive Background Monitoring and Threshold Matching

After the custom profile is saved, the iPhone’s low-power audio processor continuously samples incoming environmental noise through the device’s microphones. The live audio stream is converted into frequency data in real time and evaluated against the locally stored acoustic models.

When the live incoming audio aligns with the trained frequency, duration, and cadence profile beyond a predefined confidence threshold, iOS 15 registers a match. The system immediately suppresses false positives by filtering out transient room noise, and then fires a push notification, vibration alert, or audible ping to inform the user that the specific appliance or alarm has sounded.