So here we are, tuning and tweaking our speech recognition engine. The process involves letting people call our development systems and basically putting them through their paces, pretty straight forward, right? You would think, but we're finding out pretty quickly that people say the strangest things.

When you ask someone a yes or no question you generally expect a Yes or a No, right? Well, as it turns out, perhaps some of the time.

Now, it's important to note that all the participants were aware of what they were doing and their calls were being recorded so we could study their responses against the accuracy of our engine.

Would you like to accept this call? can get responses like:

• Yes
• Yea
• No
• Nope
• Not Right Now
• Sure
• Okay
• Absolutely
• Who is this?
• Hello? Hello?

Some of them particular entertaining like Who is this? (since the person being called knows its an automated system).

Stop Yelling at Me!

You all know this person, the one that the second they get on the phone the volume of their voice increases 300%.

Just in case you didn't know, volume is normalized within modern phone systems (such as your mobile phones) so no matter how much you yell it doesn't get any louder on the other side.

By the way, if you don't know what I'm talking about, you're the one I'm talking about.

A Study In Four Letter Words

Why do people think profanity is a fun way to test speech recognition systems?

Useful, not much. Entertaining, absolutely.