Skip to content
ML Atlas

Voice Assistants

Spoken assistants such as Siri, Alexa and Google Assistant that chain wake-word detection, speech recognition, language understanding and text-to-speech.

Also Siri · Alexa · Google Assistant · smart speaker · wake word

Fig. / How it works
Wake wordWake wordSpeech to textSpeech to textUnderstand requestUnderstand requestTake actionTake actionSpeak the replySpeak the reply
Role · kind
Real systemsFamous system
Level
Beginner

Siri on iPhone 4S (Apple, 2011); Alexa with Amazon Echo (2014); Google Assistant (2016)

You say 'Hey, what's the weather?' The device wakes up, writes down what you said, figures out what you want, looks it up and reads the answer aloud.

Where it fits

Speech and AudioSpeech and AudioAudio ClassificationAudio ClassificationLarge Language ModelLarge Language ModelSpeech RecognitionSpeech RecognitionText-to-SpeechText-to-SpeechVoice Assistants
+All relationships as a list

Is a kind of / part of

See it on the full map

How it works

Apple introduced Siri with the iPhone 4S in October 2011, the first voice assistant built into a smartphone’s operating system. Amazon’s Alexa (with the Echo speaker, 2014) and Google Assistant (2016) followed and put assistants in kitchens and cars.

Under the hood, a voice assistant is a pipeline of models:

  1. Wake word: a tiny always-on model (audio classification) listens only for “Hey Siri” or “Alexa”, on the device.
  2. Speech recognition: turns your request into text.
  3. Understanding: classic assistants matched the text to a fixed list of intents (“set_timer”, duration = 10 min). Newer ones increasingly hand the request to a large language model, which copes far better with free-form requests.
  4. Action: check the weather, play music, switch a light.
  5. Text-to-speech: read the answer aloud in a natural voice.

Some assistants also recognize who is speaking (speaker identification) to personalize answers. The always-listening microphone is why wake-word detection runs locally (edge AI) and why privacy settings matter.

Everyday use cases

  • Case 01

    How you've probably met it

    Setting a timer while cooking, asking your car for directions, or turning off the lights by voice.

Strengths, weaknesses and when to use it

Strengths

  • Hands-free and accessible for people who cannot easily use screens.

Weaknesses

  • Classic intent-based assistants were rigid; always-listening microphones raise privacy concerns.

When to use itAs the everyday example of chaining several audio and language models into one product.

Compare with

Go deeper

Sources

  1. [1]MIT Technology Review, Apple's New iPhone Would Like to Talk (2011)
  2. [2]MacRumors, Siri Turns 10 (2021)

Facts checked as of 2026-10-12