I suspect the problem people have seen with wands has to do with the Symbology used or the printing quality. In particular trying to scan anything from the grocery can be very bad with wands. There are various problems: the shape of the goods (wands work well on a flat even surface), the UPC code (it was designed for allowing great variation in printing and must be used in conjunction with a database to be reliable), and the wide variation in printing (despite detailed specifications my experience is that manufacturers can be very wishy washy in their barcode quality).

Now, if you use a symbology like Code 128 or Code 39; print it using a photographic process, on a dedicated barcode printer, or a laser printer according to specifications and have a flat surface you will always get good results.

As far as the algorithm goes, this is publish for both codes I mention above. They are available from AIM (http://www.aimglobal.org/technologies/barcode/). It includes the correct width averaging method.

We have a barcode reading product based on a 8051 type chip, so what you are trying can certainly be done, but I was wondering what was considered "too" expensive. There was a time when a laser scanner cost over $1000, but now they are plenty scanners like that for under $150. Adding speach to one of those would certainly be less work and easier to use than the roll-your-own wand approach. Of course, it would take more battery power (if you are using batteries).