Vendor page
Hey Siri: An On-device DNN-powered Voice Trigger for Apple's Personal Assistant
Apple Machine Learning Research, 2017
Read the originalRead on October 16, 2026
- What it is
- Apple's technical article on the 'Hey Siri' wake-word detector, which runs on the iPhone and Apple Watch and is the most detailed public description of Siri's speech models that we found. It explains how a deep neural network scores sound on the device and when audio reaches Siri's servers.
- How much weight it carries
- Written by Apple's own engineers, so it is authoritative for how Apple built that detector in 2017. It covers the wake-word detector, not Dictation, and describes a 2017 system that Apple has since changed, so it shows the design rather than the current model.
Facts taken from this source
- Apple's 2017 Hey Siri article says the detector uses a Deep Neural Network to turn the acoustic pattern of your voice into probabilities over speech sounds.
“uses a Deep Neural Network (DNN) to convert the acoustic pattern of your voice at each instant into a probability distribution over speech sounds” (Opening section on how the detector works)
- In Apple's 2017 Hey Siri article, audio reaches the Siri server only if the stages on the iPhone pass it on.
“If the various stages on the iPhone pass it on, the waveform arrives at the Siri server.” (Section headed Further Checks)