# What was the Amazon Echo, and how does Alexa understand what you say?

Date: 2021-07-14 · Author: Robert Scales · https://guidetoartificialintelligence.com/en/learn/what-was-amazon-echo-and-alexa/

**Answer:** On 6 November 2014, Amazon unveiled the Echo, a cylinder-shaped speaker that stays switched on and responds when you say the wake word "Alexa," with its intelligence running in Amazon's cloud. It was sold first by invitation only, and it turned the voice assistant from a phone feature into a device for the whole home.

Most of us met talking computers through our phones. Apple put Siri on the iPhone in 2011. But you had to pick up the phone, press a button and speak into it. On 6 November 2014, Amazon tried something different: a speaker you leave in the kitchen or living room that is always ready to listen, from across the room, without touching anything.

That speaker was the **Echo**, and the voice inside it was **Alexa**. Today, in mid-2021, Alexa runs on a huge range of devices, and the idea of simply talking to a box in the corner no longer sounds odd.

## What Amazon announced

Amazon's product page for the launch described a tall cylinder with a ring of light on top. Its pitch was simple: ask for information, music, news or the weather, and get an answer instantly. The device started working when it heard the wake word, "Alexa."

At launch, Echo could:

- play music from Amazon Music, Prime Music, iHeartRadio and TuneIn, or stream other services from a phone over Bluetooth;
- read news and weather, including local radio stations, NPR and ESPN;
- set alarms and timers and keep shopping and to-do lists;
- answer common questions using sources such as Wikipedia.

The price was $199, but Echo was sold **by invitation only**. Customers requested an invitation, and Prime members who received one could buy it for $99 for a limited time. A companion app on Fire OS, Android and web browsers handled setup, alarms and lists.

The invitation period lasted about seven months. On 23 June 2015, Amazon announced from Seattle that any customer in the United States could now order Echo for $179.99, with shipping from 14 July. By then, the company said, Alexa had learned dozens of new features, including Pandora, sports scores and smart-home controls, and outside developers were testing a free kit for building new "skills."

## Who built it, and where the name came from

The Echo hardware came from Amazon Lab126, the company's in-house device lab. Alexa's software grew partly out of two companies Amazon had bought: Evi, a British question-answering service acquired in 2012, and Ivona, a Polish speech synthesizer acquired in 2013. Evi supplied the ability to answer questions; Ivona supplied the voice.

Amazon's developers have said they picked "Alexa" partly because of its hard "x" sound, which makes it easier for a machine to pick out of ordinary conversation, and partly because it calls to mind the ancient Library of Alexandria.

## How a voice request works

Echo's main trick was hearing you from across a noisy room. Under its light ring sit seven microphones. Using a technique called **beam-forming**, the device combines their signals to focus on the direction your voice comes from, and noise cancellation lets it hear you even while it is playing music. This is called **far-field** voice recognition, as opposed to talking right into a phone.

The heavy lifting happens elsewhere. Amazon described Echo's "brain" as living in the cloud, on Amazon Web Services, where it could keep improving and adapt to each household's way of speaking. Here is the journey of a simple request:

| Step | What happens | Where |
|---|---|---|
| 1 | You say "Alexa, what's the weather today?" | your kitchen |
| 2 | The device notices the wake word and starts sending audio | on the Echo |
| 3 | **Speech recognition** turns the sound into text | Amazon's cloud |
| 4 | **Language understanding** works out the intent (weather) and details (today, your city) | Amazon's cloud |
| 5 | A weather service fetches the forecast | Amazon's cloud |
| 6 | **Text-to-speech** turns the answer into a spoken sentence | Amazon's cloud |
| 7 | Echo plays the answer through its speaker | your kitchen |

All of this takes a second or two. Each step relies on machine learning, models trained on large amounts of recorded speech and text, which is why Amazon could promise that the device would get "smarter" without anyone buying new hardware.

> **Try it:** Next time you use any voice assistant, say your request in three different ways ("weather today," "do I need an umbrella," "is it going to rain"). Note which phrasings it understands. You are testing step 4, the intent-matching part.

## Why it mattered

Echo created a new category: the smart speaker. Amazon opened Alexa to outside developers through the Alexa Skills Kit, and the number of skills grew from about 1,000 in June 2016 to more than 90,000 by April 2019. Other manufacturers could build Alexa into their own products, and in late 2016 Amazon began offering the speech and language technology behind it to developers as a cloud service called Amazon Lex. In January 2019, Amazon said it had sold more than 100 million Alexa-enabled devices.

Echo also raised a lasting question. A device with microphones that is always on, in the middle of the home, makes people uneasy about what is recorded and who hears it. Amazon's answer is that devices only stream audio to the cloud after hearing the wake word, but privacy concerns have followed smart speakers ever since.

## How is Alexa different from a language model like GPT-3?

Alexa is built mostly around **intents**: a request is matched to a known task, like checking the weather or adding milk to a list, and a specific skill handles it. That makes it reliable for short, practical commands, but it can stumble on open-ended conversation or a question no skill was built for.

A large language model such as OpenAI's GPT-3, released in 2020, works the other way around. It has no list of tasks; it simply predicts what text should come next, which lets it write, summarize and chat on almost any topic. The price is that it can produce fluent answers that are wrong, and it has no built-in connection to your calendar, lights or shopping list.

Many researchers expect future assistants to combine the two: the open conversation of a language model with the dependable actions of a voice assistant. For now, in mid-2021, they remain largely separate tools.

## Sources

- Amazon Echo - Official site - Request an invitation (archived 7 November 2014): https://web.archive.org/web/20141107234621/http://www.amazon.com/oc/echo/
- Amazon Echo Now Available to All Customers: https://press.aboutamazon.com/2015/6/amazon-echo-now-available-to-all-customers
- Amazon Alexa: https://en.wikipedia.org/w/index.php?title=Amazon_Alexa&oldid=1033404336