Voice assistants and smart speakers: Privacy and technology

In short: Voice assistants wait for a wake word, then send speech to servers for analysis and carry out commands; the microphone and data storage deserve attention.

In more detail: Alexa, Google Assistant, Siri and other assistants control music, lights, timers and information by voice. Smart speakers have microphones, loudspeakers and an internet connection for this.

In Depth

How it works

1. The device constantly listens locally to detect the wake word (“Alexa”, “Hey Google”).
2. Only then is audio recorded and sent to the cloud.
3. There speech is recognised (ASR), intent detected (NLU) and a reply generated (TTS).
4. The device carries out the action. Newer devices also process simple commands locally.

Privacy risks

• False activations by similar-sounding words send conversations to the provider unintentionally.
• Recordings are stored and partly listened to by humans for quality checks (depending on settings).
• Linking with accounts, purchases and smart-home devices produces a detailed usage profile.
• Insecure devices can be hacked; AI imitates voices (voice cloning).

Protection

• Use the mute button when confidential conversations take place.
• In the app review and delete recordings, enable automatic deletion, switch off use for training.
• Protect voice purchases with a PIN.
• Do not place the device in the bedroom or an office with confidential content.
• Update firmware, secure Wi-Fi, a separate guest network for smart-home devices.
• Inform guests (they have a right not to be recorded unintentionally).

Alternatives

Local systems (Home Assistant with Wyoming/Whisper, Rhasspy) keep speech in your own network but need setup and are less convenient. Check whether you really need the convenience: a button or an app often does the job too.

See also: Speech recognition and synthesis, Microphone, Data protection in everyday life