You will remember the demonstration at our January R&D Committee
meeting of BARTALK, the program and hardware out of Kentucky that
reads the UPC barcode on products and then speaks the identity of
the product. The demo version contained just a few entries in its
database, primarily chosen from the line of merchandise sold
through vending stands. This choice was made by the Kentucky
Department for the Blind and Mr. Wayne Thompson, their engineer.
This is to bring you up to date on the work that has followed the
February demo.
NFB is exploring the potential of expanding the system to the point
where it would be of value to blind homemakers nationally. To this
end, I have obtained the UPC data base for national brands sold
through super markets. There are ninety-one thousand entries in
this database, which also includes products sold as local brands
through Kentucky supermarkets.
I have engaged Mr. Thompson to massage this database down to size,
(it was fourteen megs before we worked on it, and now it is just
3.4 megs. Of course, we have kept all of the data in tact as the
file was worked over.
A new program must be written that will efficiently retrieve
information from this rather large database. The one demo'ed at
our meeting, won't do. In a test run using the original software,
it took 21 seconds for the program to locate the identity of the
product at the end of the database.
To speed things up, we are testing binary search, interpolative
search and hashing algorithms. We have to get response time below
two seconds using a slow microprocessor, like the 80286. This is
dictated by the fact that the cost must be contained, if a lot of
people are to get to use the program.
I'll fill in a lot of details when we meet in Chicago, either on
the Committee agenda or in the hall by the elevator.
This archive was generated by hypermail 2b29 : Sun Dec 02 2012 - 01:30:03 PST