Easier Said Than Done

Ever watch any of the Star Trek shows and wish your computer could operate on voice command like that? Speech recognition is one of those technologies that's been under development for years, and for its advances, well, it still stinks.

Most of the problem is in the software, no question. A lot of very smart people have worked very hard on this and it's still not perfect: not even close. Plus, most of these systems require you to wear a headset, since the microphones were so poor that they had to all but stick the mic in your mouth to pick up anything. It reflects the low quality of the microphones. Not to mention it's annoying to wear those things.

SoundMAX and Andrea Electronics have teamed up to alleviate some of this mess. SoundMAX, an Analog Devices company, makes audio chips for PC motherboard. Its CODECs offer Dolby Digital 5.1 sound quality, providing far better sound output than your standard AC97 audio chip. Andrea makes microphones and software to optimize speech recognition and voice command software.

Now, the SoundMAX chips offer beautiful, digital sound. As most hardcore gamers know, if your motherboard comes with on-board audio, the first thing you do is throw in a Sound Blaster and disable the onboard audio in the BIOS, because that AC97 audio is terrible. SoundMAX Cadenza is every bit as rich in sound as the Sound Blaster Audigy, but what I really wanted to test out was the speech recognition.

Together, these two products make a very good team for speech input and command of a computer. I was sent a PC with the SoundMAX chip on the motherboard and an Andrea Superbeam Array microphone to try out, after seeing it demonstrated at the previous Intel Developer Forum.

I was impressed with its clarity at voice command recognition, and the fact that you didn't need to wear a headset. The unit actually has two microphones built in, as the picture on the above link shows, and sits on top of your monitor. Now that's a good foot or two away, unlike the one or two inches of space between a headset microphone and your mouth.

The microphone is nice, but the real value is its PureAudio noise cancellation technology, which allows the microphone to sit two feet away and still work. Since I'm an apartment dweller, there's all manner of noise around here, so the microphone and software got a real test.

First off, you need to spend about 30 minutes tuning the software, so it gets used to your voice. I used the speech recognition software in Microsoft's Plus! for Windows XP, but you can use any package, such as Dragon NaturallySpeaking. The microphone and chips are independent of the voice command software.

The tuning was done in silence. Then I tortured the system. The door and window were opened just as the landscapers showed up, the radio was turned on, and it was show time.

Microsoft Office XP has some pretty comprehensive voice command/speech recognition technology, which is where I had the most fun. Even with a leaf blower and the radio going, the Andrea microphone operated beautifully while running Office. Sitting a foot from the monitor, all I had to do was say "select all," "format," "paragraph," "line spacing," and "double," and voila, my Word document was double-spaced.

Despite the sounds of man, nature and my radio, and a distance from my mouth, the microphone picked up almost everything and I rarely had to repeat myself. That's a major improvement over anything I've used in the past.

I can't say the same for Microsoft Office's speech recognition. This is why I differentiate between voice command and speech recognition. A friend once joked that speech recognition software is tuned to a perfect Midwestern cadence, so unless you speak like Dan Rather, you're probably going to have a rough time using the software. I found this to be true in the speech recognition. My careful dictation and attempts to suppress my northeastern accent not withstanding, what came out was a bigger mangling of the English language than George W. Bush at his finest.

What was said: I wonder if I should save this and run it in my column to show how bad the speech recognition is in Word.
What came out: To book a bouquet this is said a should run this in my column show over the speech recognition is in word.

What was said: Oh my God this stinks
What came out: all my daddy stake

The sound of me laughing was recorded as and and and and and and and and and

It wasn't always a disaster, though.

What was said: The quick brown fox jumped over the lazy dogs
What came out: The quick brown fox jumped over the lazy dog's

What was said: Obviously Microsoft's voice recognition software needs more work
What came out: Obviously Microsoft's voice recognition software needs more work

Ironic, isn't it? A slag on Microsoft was recognized perfectly.

I've been looking for decent voice command software for a friend who is in the advanced stages of muscular dystrophy; his arms have almost completely failed him by now. Soon he won't be able to operate a computer. He plays EverQuest, but there's no software that will let him operate the game by voice, and unfortunately, the Superbeam microphone and SoundMAX chip won't help until the EverQuest software itself is voice-enabled.

I think a lot of companies have eschewed voice command technology because it's been so bad. Why spend the time and money on voice commands for your software when it will barely work because the input is so weak?

One thing is clear: the Andrea microphone and SoundMax chip do a great job receiving audio input and commanding applications like Office or Windows Media Player, which is also voice command enabled. Despite the racket that I allowed to infiltrate my apartment, I could operate Word and WMP without ever touching the keyboard.

Perhaps improvements in microphone and noise cancellation by Andrea and SoundMAX will help change some minds at ISVs and make them reconsider voice command functionality. I'm willing to bet all these people need to see is that it can be done on the hardware side.