🎙️ Speech and Voice Assistants

3rd Grade 🤖 Artificial Intelligence

A voice assistant is a program you can talk to out loud: ask for a song, ask about the weather or set a timer. Behind the scenes there are several steps of AI — turning sound into words, guessing what you asked for, and turning the answer back into sound.

Free · US Letter · no sign-up
Google Classroom Save to Pinterest Email me

Step one: from sound to words

When you talk, the air vibrates, and the microphone turns those vibrations into numbers. To a computer, a spoken sentence is a long wave of numbers going up and down. The first job is to figure out which words are hiding in that wave. That’s called “speech recognition.”

It’s an especially hard job: everyone talks differently — fast or slow, in a high or low voice, with a different accent. There’s background noise, people talk at the same time, and some words sound almost exactly alike. Speech recognition programs learn from thousands of hours of recordings where people wrote down what was said, and they find patterns between sounds and words.

Step two: what did you actually ask for?

Once the words are written down, the request has to be understood. “Play a happy song,” “Put on some cheerful music” and “I want to hear something upbeat” are different sentences with a similar request. The program learned from lots of examples to spot the intent — play music — and the details — happy.

Remember that here too, it doesn’t “understand” like a friend does. It guesses the most likely intent based on patterns. So an unusual sentence, a joke or a complicated request can confuse it, and it might do something you didn’t want or answer “Sorry, I didn’t get that.”

Step three: from the answer to sound

When the answer is ready, the program has to “talk.” This is called text-to-speech — turning text into spoken words. Computer voices used to sound robotic and choppy. Today, by learning from huge numbers of recordings of human voices, they can sound very natural.

This helps lots of people: someone who has trouble seeing can hear what’s written on the screen, and someone who has trouble typing can dictate messages. But a natural-sounding voice doesn’t mean there’s someone behind it who feels or understands. It’s still a program.

The wake word and privacy

Most voice assistants wait for a “wake word” — a special word or phrase — before they start listening to a request. That means the device is listening all the time just to catch that word. Sometimes it hears something similar by mistake and wakes up without being called.

Because these are recordings of voices inside your home, privacy really matters. Ask your parents to check the device settings: which recordings are saved and how to delete them. Simple rules for kids: don’t tell a voice assistant your full name, address, password or phone number, don’t ask it to buy things without a parent, and if it says something strange or scary — tell an adult. One more little rule: a voice assistant isn’t the place for secrets, because it’s a device connected to the internet, not a friend who promises to keep them.

Why it gets confused sometimes

Voice assistants have a harder time when there’s noise, when you talk very fast, when you mix two languages in one sentence, or when you speak with an accent that didn’t show up enough in the recordings they learned from. Kids’ voices are sometimes hard for them too, because the recordings they learned from were mostly of adults. If you speak slowly and clearly and keep your request short, the chance it understands you goes up.

You can test this with a small experiment with a parent: ask the dictation tool on a phone to write the same sentence once in a quiet room while speaking slowly, and once quickly with music playing in the background. Compare the two results. Count how many words came out wrong each time, and try to guess why those words in particular. That way you see with your own eyes how noise and speed affect speech recognition.

Key facts

  • Speech recognition turns a sound wave into written words.
  • A voice assistant guesses the intent of a request from patterns.
  • Text-to-speech helps people who have trouble reading or seeing.
  • Most assistants wait for a wake word before they listen to a request.
  • Don’t give a voice assistant personal details or passwords.

Quick quiz

  1. What does speech recognition do?

  2. Why is speech recognition a hard job?

  3. What is a “wake word”?

  4. Does a voice assistant with a natural voice feel anything?

  5. What should you never tell a voice assistant?

  6. How can you raise the chance the assistant understands you?

  7. How does text-to-speech help?

  8. What should you do if a voice assistant says something scary?

💭 Think and talk about it

  1. Try asking for the same thing in three different ways. What does the voice assistant need to notice to understand it’s the same request?
  2. Why do you think kids’ voices are harder for some voice assistants?
  3. What rules would you set at home for using a voice assistant?

Questions kids ask

Is a voice assistant listening all the time?

Most assistants are listening all the time only to catch the wake word, and they start listening to the request after it. Sometimes they wake up by mistake. Check the settings with a parent.

Why does the voice assistant give a wrong answer?

It guesses from patterns — both the words and the intent. If it heard wrong or guessed wrong, the answer will be wrong, so check important answers.

Can a voice assistant understand languages other than English?

Many assistants and programs support lots of languages, like Spanish, French or Chinese. As with any language, they work better when you speak clearly and in simple sentences.

🔎 Ask Thinkapedia another question