I just hung up the phone on a speech recognition program. I know I shouldn't have done it (bein' as how that's my career field an' all). My parents taught me not to hang up on anyone in the middle of a conversation. I assume that means anyTHING as well. But in my defense, from the program's first prompt, it was obvious that whoever wrote the program just did a wholesale replacement of speech for Touch Tone.

That really frosts me.

Sure, maybe the program is just a placeholder until they get the real speech application built. Maybe the programmer is new to speech app development and just needed to get used to the technology, so he figured, "We've got our legacy DTMF program sitting here. Instead of having people press 1 or 2, I'll have them say 'one' or 'two.' That'll be a safe way to figure out what I'm doing, and it'll buy me some time so I can build an app that really takes advantage of the power of speech recognition."

Maybe.

But having dealt with this particular company before, it's more likely that the programmer was ordered to "put out a speech program as quickly as possible regardless of whether it helps the caller or not. Streamlining we don't care about. Flatten the menu structure? Why bother? Just get the thing built, like, yesterday, so we can claim that we're part of the speech world. We'll send out the press release, and you get to keep your job. Capisci?"

My friends, we can do better
But speech apps should be better than that. When giving my two-minute spiel on the value of speech technology, I often cite its "flatness" potential. That is, instead of requiring the caller to step through miles of menu layers, a good speech recognition app can save you a good deal of time and frustration.

Consider this clunky menu-driven voice app (the name of the business has been changed to protect the guilty):

MACHINE: "Welcome to Circeo's Auto Park. For Sales, press or say 'one.' For Service, press or say 'two.' For Parts, press or say 'three.'"
CUSTOMER: "Two"
MACHINE: "You chose two for Service. For Domestic Service, say 'domestic service.' For Foreign Service, say 'foreign service.'
CUSTOMER: "Foreign service."
MACHINE: "You chose Foreign Service. If you are calling to set up an appointment, say 'appointment.' If your car is currently being worked on, say 'current work item.'"
CUSTOMER: "Current work item."
MACHINE: "Do you have a work order number? Say 'yes' if you have a work order number, or 'no' if you do not have a work order number."
CUSTOMER: "Yes."
MACHINE: "What is your work order number? Please speak slowly and distinctly."
CUSTOMER: "4-5-5-7-2"
MACHINE: "You said, '4-5-7-2.' If this is correct, say 'yes.' If this is incorrect, say 'no.'"
CUSTOMER: "No."
MACHINE: "What is your work order number? Please speak slowly and distinctly."
CUSTOMER: "4-5-5-7-2"
MACHINE: "You said, '4-5-9-7-2.' If this is correct, say 'yes.' If this is incorrect, say 'no.'"
CUSTOMER: "No."
MACHINE: "What is your work order number? Please speak slowly and distinctly."
CUSTOMER: "4-5-5-7-2"
MACHINE: "You said, '4-9-9-7-2.' If this is correct, say 'yes.' If this is incorrect, say 'no.'"
CUSTOMER: "NO! YOU %*#*&@ !!"

...and so on, and so forth.

Now how much better would it be if the app developer put a little time and effort into the call flow?

MACHINE: "Welcome to Circeo's Auto Park. How can we help you today?"
CUSTOMER: "Well, I left my car there at 10 this morning and I'm wondering when it'll be ready."
MACHINE: "Certainly, do you have a work order number?"
CUSTOMER: "Uh, yeah, it's 45572."
MACHINE: "Thanks...I see that your car is right on schedule and that you can pick it up any time after 3 this afternoon. Is there anything else we can do for you?"
CUSTOMER: "Nope. That's it. Thanks."

Not bad, eh? The first example shows a customer getting extremely frustrated talking to little more than a Lost in Space robot. The second example is not only smoother and more effective, it cuts the customer's time on the phone by more than 70%. Now if you've already guessed it, the first example is real and the second example isn't. In fact, I've yet to personally encounter such an efficient speech app as the second one. It would have to be a fairly sophisticated program to pick out keywords in the customer's first sentence and correctly identify the customer's issue.

But that's the idea: speech is different. It's versatile, and a good speech dev treats it that way. That usually means a much flatter program than a traditional DTMF app. Not just for the sake of flattening, mind you. After all, flat doesn't necessarily equate to efficient. But just as I was taught in college to "write tight, and expand only where necessary," we should start each speech app design by trying to "think flat, and deepen only where necessary." If we do, we'll help to advance the technology, cut down on call times and service costs, and consumers the world over will thank us for it.

That is, if their parents taught them right.

Cup o' Joe
"Just got this in today," said Phil the Coffee Junkie, as he opened the fresh package of Delta Ouro. "Check out those beans. Pretty big, and the roast looks darker than the last batch they sent me." I never know what Phil is talking about. He's got this coffee business down cold, and I'm pretty sure I'll never even approach his level of coffee knowledge or appreciation. But after Phil poured me a cup of pure Delta Ouro, even I could see that the deep mahogany color of the bubbles somehow held more flavor than the usual blend. Maybe it was the frost on my window pane. Maybe it was the unusual melancholoy that had accompanied me most of the morning. But this Delta Gold hit me just right. Rich, smooth, mellifluous. It was the perfect companion on a cold winter's day in The Great Northwest.