How does a voice assistant work?

The European Data Protection Board’s guidelines on virtual voice assistants describe one as software that provides spoken dialogue with a user in natural language, and set out the steps a request goes through:

  • On standby, it listens on the device only for its wake-up expression, keeping a few seconds of audio in a buffer.
  • When it hears the expression, it opens a listening channel and sends the request to the provider, where the speech is transcribed.
  • Natural language processing extracts what the user wants, and a dialogue manager decides how to respond.
  • It builds an answer, often spoken through text to speech, or sends an action to a device, such as playing music or changing the heating, then returns to standby.
Sources

How is a voice assistant different from a voice agent?

The terms overlap and are used loosely. A general voice assistant works for its user: it lives on their phone or speaker, can connect to their accounts and handles many kinds of task. A voice agent usually works for a business: it answers that business’s customers about its products, on a phone line or a website, from information the business provides.

Example: “What’s on my calendar tomorrow?” is a job for a personal assistant. “Does your team plan include single sign-on?”, asked on a software company’s pricing page, is a job for that company’s voice agent.

Listening differs too. An assistant that waits for a wake word is always listening on the device, which is why the EDPB guidelines address accidental activations and other people’s voices caught in recordings. A website voice agent hears nothing until the visitor starts a call and allows the microphone.

How does Kawl Now compare with a voice assistant?

Kawl Now is a website voice agent, a narrow kind of voice assistant that a SaaS company puts on its own website. It answers questions about that company’s product from knowledge the company has approved, in a widget on the site.

It has no wake word and never listens in the background. The microphone opens only after a visitor clicks the widget, types what they need and allows access, and the call ends when they hang up or at 10 minutes.

It doesn’t carry out tasks for the caller: it can’t see or change their account, book meetings or control devices. When asked, it records a bug report or feature request, and when it can’t help, it says so and points the caller to another way to reach your team.

Common questions

Is a voice assistant always listening?

Most phone and smart-speaker assistants listen on the device for their wake word, keeping only a few seconds of audio until they hear it, as the European Data Protection Board describes. A website voice agent such as Kawl Now hears nothing until the visitor starts a call and allows the microphone.

What is an AI voice assistant for a website?

It’s a voice assistant that lives on a company’s website and answers visitors’ questions aloud about that company’s products. Also called a website voice agent, it runs in the browser through a widget rather than on a phone line.