I can’t understand a word you’re saying…

Land Rover 109 lwb 1980

Land Rover 109 lwb 1980 (Photo credit: Wikipedia)

Whenever I see those adverts for the iPhone where they show off the speech recognition software (called Siri), I can’t help a wry smile forming on my face. I was asked to undertake a feasibility study of speech recognition software once. It was roughly 15 years ago in the Pentium era.

My boss thought that the technology would prove useful in our care control system. The elderly and the infirm wear electronic pendants and when they get into trouble, they simply press the button on the pendant. The base station under the phone wakes up and dials the control centre so that the person in trouble can be connected an operator.

Local authorities across the land had signed up for the software and it was a big success. In metropolitan areas such as Birmingham, they had pendant wearers from all kinds of ethnicities and many of them could not speak English. It was impractical to teach all the control centre operators all the different languages so they relied on interpreters which had to be called in on demand. This system was very expensive and there was an inevitable delay in getting relief to where it was needed.

If we had speech recognition built into the software, we could connect the pendant wearer directly to the right interpreter. We could also weed out all the false positives like people who pressed the button by accident. Sounds perfect – all we need is speech recognition.

I installed the Lernout & Huaspe speech recognition engine, wrote a simple little program to test out the software and plugged in my microphone. On the screen was a list of three cars. When I spoke into the Microphone, all the software had to do was work out which of the three cars I called out (what can I say – I like cars).

The first attempt went well; “Land Rover” I said carefully into the Microphone. A message popped up saying “You said Land Rover”. So far so good.

Second attempt; “Mini”. Again, a message popped up saying “You said Land Rover”. Not so good.

I tried saying “Mini” again, speaking more slowly and clearly this time, it still thought I was saying Land Rover.

I tried saying it quickly,

I varied the tone of my voice.

I tried different volume levels.

I even tried a different Microphone; “You said Land Rover”.

I thought it might just be the word “Mini”, so I tried “Jaguar”; “You said Land Rover”.

I swore at it. “You said Land Rover”.

The idea was doomed. If it couldn’t recognise me speaking clearly into a Microphone an inch away from my mouth with no background noise – it was never, ever going to work in our target environment. Most of the pendant wearers were totally deaf, so the TV would be on at maximum volume and they would never have the good grace to fall over right next door to the base station so their voice would be distorted by distance.

I went back to my boss and told him that it would never catch on.