Instruction
(referring to the pages which follow the instructions below):
When
the start button is depressed, the first fragment of a word is presented.
In this case the first sound of the word. Your task is to guess
what word it is and write your answer in the response window below
the button. When the 'Next' button is depressed a slightly greater
fragment of the same word is presented, and so on. The point here
is to show that most words are recognised much before we hear the
complete word. Common words may be recognised as early as only a
third or so into the word. This shows that the speech recognition
process is at least partly 'top down', that is based on predictions
we make based upon fragmentary acoustic information.
(The particular
word used in this demo is a Swedish word, but a very common one
that occurs in a similar shape in most western languages.)
In this demo,
it does not actually matter what you enter in the response window
and the results presented are not based on your own responses. The
component that computes the results has been excluded from this
demo for practical reasons.
The 'Send in'
and 'Group result' buttons do not work in this demo.
Click here
to start the demo!
|