DSP Robotics Support

Hello all,
Is there any way on earth to count words in a audio stream? Lets say I sang the words, "marry had a little lamb"
The count would be 5 words or if I just spoke the words, " 1 2 3", the count would be 3 and so on and so on.

Thanks to who ever has an idea!
Later then, BobF.....

Maybe defining a threshold and a time interval, then calculating RMS, and if previous and current RMS differ by at least a certain amount, count +1? May also count syllables if spoken or sung very emphasised, and could be counting 0 if spoken or sung like a robot. But that's how I would give it a try.

I think that such a system proposed by tulamide would require the singer to break up each word in a staccato fashion since the schematic would have no intelligence to recognise real words.
With feedback from the schematic in some form the singer could quickly train the technique to get the required result.
Alternatively, a Ruby expert could create a complete voice/word recognition system, given a few years

Cheers

Spogg

it is actually much more complicated then it seems. Speech recognition is one of the hardest computer problems. Sound-wise there is no real definition of a word. Gap detection (as proposed by tulamide above) works for articulated speech maybe 70% but then there are words that sound-wise meld with the previous or next word (e.g. "a water" is pronounced closer to "eiwoder" than "ei woder") and a "gap" is also technically a constant (for example in a word "o-ou"). For legato singing the words tend to merge completely.

Hey gang,
well thanks so much for the input, I knew it would NOT be easy. Think I will abandoned this project for a while.
On to my next crazy idea I guess. Again Thanks!
Later then BobF.....