On 4 March, I came across an article on the BBC News web site reporting that some researchers at Stamford University and at Edinburgh University found, Computers find male voices ‘harder to recognise’. Now I am male and as I cannot hand write, I have been using voice recognition technology (VRT) since I was 10 years old. So, I “screen read” the article to see what insights it had for me.
When I started using VRT, I could barely read but had to read “training text” for the computer to learn my voice. My mother had to sit behind me, whispering the text into my ear in short phrases, which I repeated into the microphone. I was using Dragon Naturally Speaking version 5. It was hard work, but I persevered as this was the last chance saloon for me as I was unable to write and typing was painfully (literally!) slow. I adopted the “good practices” recommended by the software supplier, such as correcting any recognition errors as I went along using the “Correct That” function, rather than overtyping. The result was I achieved a recognition level of over 98%, talking in a natural way, at normal talking speed.
So, what about the Edinburgh and Stamford Universities’ research? The BBC report says:
Computers failed to understand men’s speech because they make “umm” and “err” sounds more frequently.
Comment: I could have told them this ten years ago! Of course sloppy speech makes recognition levels bad and perhaps men umm and err more than women! People with speech ticks such as saying “y’know” in every sentence also have problems. Isn’t there and old computing saying of “trash in, trash out”?
Computers made mistakes with words which sound similar and can occur in similar contexts, such as “I saw him” or “I saw them”.
Comment: This is a good point. They ought to try “I scream” and “ice cream”. In one of my Religious Studies essays I had “cheeses of Nazareth” leading his disciples. I have however got round this now and I’ll tell you how in a later blog.
Variations in pitch, tone and speed can also cause the system to misunderstand voices.
Comment: Not for me; even though my voice broke, my good practices carried me through and I can talk as fast I want to.
It then said the research was to “improve the accuracy of automated ticket booking lines.” Ah!!! This research was not for people like me after all. It was all about replacing call centre staff with machines; this software would have to deal with any voice without the benefit of any training! Perhaps, if they crack this problem, it will eventually benefit people who cannot hand write or maybe it will be so expensive, it will remain the preserve of the major corporates.
Only the future will tell; until then, I’ll keep using my Dragon Naturally Speaking software coupled with some common sense good practices.
