Men’s voices harder to understand for speech recognition

In a BBC news story recently titled “Computers find male voices ‘harder to recognise’” describes how Edinburgh University scientists found that computer speech recognition software finds male voices harder to understand than female voices.

This they attribute to the fact that men “make ‘umm’ and ‘err’ sounds more frequently.”

I can attest to this. It is a common problem in general, and I was guilty of doing this in presentations in between lines as I paused to catch a breath. It’s also common when pausing between words when using voice recognition. A bad habit? Perhaps, but a common habit nonetheless.

In addition, the scientists found that, “Variations in pitch, tone and speed can also cause the system to misunderstand voices” – which goes without saying. No two people talk in the same way or style. If we did, we’d be a very boring species! Dialects and regional variations also come into play, and as such I can certainly see why speech recognition software has its work cut out to achieve results.

My own experience with trying out speech recognition wasn’t great. I think my low toned Scottish accent with plentiful doses of ‘umm’ and ‘err’ didn’t help either, but I did feel a bit aggrieved at its inability to learn my accent. It was some time ago so the software may well have changed and improved a lot since then, but I remember feeling frustrated at the many hours I had sunk into the system hoping it would learn my voice, all for naught in the end. Voice recognition in my mind has always had great potential for relief of over-used arm and hand muscles, but the effort involved in learning how to use the software, ie teaching the software to recognise your voice, for was a soul destroying and futile task!

It’s a sad but true fact that for me there really is no comparable alternative to the keyboard in this day and age!

Rate this post! 1 Star2 Stars3 Stars4 Stars5 Stars (No Ratings Yet)

Loading...

2 comments

  1. Adrian Kelly says:

    Hi. I read your article with interest, but would, respectfully, point out that it sounds rather dated. For a start, Dragon Naturally Speaking 10.1 – the latest version – is remarkably quick to learn your accent and for general use does not require much training at all; it can be honed for dialects; is remarkably accurate. So, I really suggest you get the latest version of this software. I think you will be surprised to see how far it has advanced since your, apparently, first use of it.
    Best wishes,
    Adrian

  2. alanf says:

    Hi Adrian,

    You are partly correct, my encounter with speech recognition was 4 years ago now and I expect the software has perhaps improved a bit by now. However the reserach quoted in the article about male voices being harder to recognise was from a March 2010 article quoting presumably recent research done by Edinburgh University, so I’d say that whatever they found isn’t ‘dated’ unless of course they are using outdated software too!

    Alan

Leave a Reply

Note: Comment moderation is in place, please don't submit twice.

*