Five tips for accessible speech recognition using AI

Leonie Watson of WC3 The Amazon Echo and Google Home have revolutionised speech-enabled devices and made smart homes a reality.

Leonie Watson from the W3C Consortium offered top tips for accessible conversations at TechShare Pro 2017. In addition, Watson spoke about the legends preceding today’s voice-enabled machines.

See 5 tips for accessible, conversational interfaces

A brief history of artificial speech

“As far back 1000 years ago, people were thinking about the concept of artificial speech,” says Watson, director of developer communications for the Paciello Group.

The Paciello Group is a member of the W3C (World Wide Web Consortium) Advisory Board, and co-chair of the W3C Web Platform Working Group.

Legends of conversing “machines” date back to Pope Sylvester II (950–1003 AD).  According to an early essay, Sylvester created a “basic first dialog system,” which included speech recognition components.

Steve Jobs introduces the first talking Apple Mac

Fast forward to the 1980s, when the first text-to-speech synthesiser by DECtalk was introduced. 1984 also saw Steve Jobs demonstrate the Apple Mac’s talking capabilities for the first time with Apple MacInTalk.

“There’s been some good marketing around such technology”, said Watson, who has sight loss. “But I’ve found that talking to tech has been a laborious process - with a person having to speak very, very clearly and with specific phrases for machines to understand. Even then, the interaction has ended up bearing little resemblance to an actual conversation”, said Watson.

How Siri accelerated speech interfaces

“The thing that really changed that was Siri in 2011. For the first time, we could have something that felt a lot more like a conversation with technology. In 2014 the Windows Cortana launch followed, giving us another digital assistant that would talk back to us.”

“The same year, with the Amazon Echo, we started to see digital assistants be able to do practical things around the house, but we still needed very structured language and to ask very carefully phrased demands to get it to do things,” explained Watson. “A further leap forward came in 2015 with Google making its technology more context-aware. Meaning, for example, if a song was playing, you could ask your Google device ‘where’s this artist from? Or what’s his real name?” without having to specifically state who you were talking about.”

5 tips for accessible conversational interfaces

Watson laid out five ways that developers could make interactions with machines as clear as possible for a wide range of people.

1. Keep machine language simple

  • Think about the context of how people might be using the device. They might be driving or cooking and need short, simple interactions.
  • Offer choices but not too many choices.
  • Suggest obvious pointers to useful information.

2 Avoid idioms and colloquialisms

  • ie, terms like “it’s raining cats and dogs” or “coffee to go” might only be understood by certain audiences and so lack inclusivity.

Amazon echo3 Avoid dead-ends in conversation

  • Give the users cues around what to say or ask next to get what they need.

4 Use familiar, natural language.

  • Ie for time, say ‘three-thirty in the morning’ for the UK or US audience. Don’t say ‘zero, three, three zero a.m’.

5 Provide a comparable experience

  • Users of such technology will generally require speech and hearing to talk to machines.
  • For those with hearing loss, conversational transcripts could be posted on-screen.
  • For those without speech, the only obvious option at the moment is using simulated speech, like Stephen Hawking does, for example.

Learn more