Speech Recognition Technology

Written by Michael O'Brien
Bookmark and Share

As ashamed as I am to admit it, I have prejudices against speech recognition technology. You see, I used to be head audio engineer for a company that used speech recognition technology to try and pioneer a new way of accessing the Internet over the phone. My job was to create the audio prompts and recorded content that the service used to respond to user input.

So why did I have such a problem with the voice recognition technology our service used? The simple answer is that it just ever seemed to understand things very well. It had a specific list of commands it could understand, and if you strayed from them, you'd get lost. The learning curve was extremely steep as well.

Additionally, because we were using recorded audio prompts to speak to people, we had to anticipate every response that may have been spoken so that we could record the appropriate phrase or word and have it ready to play. Even with thousands of voice prompts recorded and a text to speech engine at work to fill in the gaps, the technology still fell short.

Necessary Improvements in Speech Recognition Technology

As time passed, however, the speech recognition technology improved. The more people that spoke to it, the better it became at interpreting commands. Over time, we also learned to better anticipate user responses and were able to create more audio prompts to deal with as many possibilities as we could. Today you can pretty much speak very naturally to voice recognition software and have it understand you perfectly. I still speak very slowly to voice interactive phone menus, however. I guess old habits are hard to break.


Bookmark and Share