I Speak, therefore I am by Hatim Kantawalla

Test Process
Voice recognition software has always been given closer scrutiny since a lot is demanded and expected from them. Accuracy is of utmost importance along with the time it takes for recognition. It is only continuous and narrative type of speech recognition which is considered to be useful. The days of discreet speech recognition programs are over and hence, all the tests were carried out without any pauses in between.

Added to these challenges are the enormous variance that already exists among individual human speech patterns, pitch, rate and inflection. These variations are an extraordinary test of the flexibility of any program. Naturally, the products were put through a stringent test which tested all the features.

Test Bench
The Test Setup consisted of an average system comprising an Intel Pentium III 500 MHz coupled with 128 MB of PC-100 SDRAM and a 8.4 GB Seagate U8 hard disk drive on an Intel SE-440BX2 platform. The soundcard used was the Creative Sound Blaster Live! Value.

Accuracy
The test for accuracy in predicting words and the way the software learns the style, accent and phonetics was done through the narration of a carefully prepared paragraph. This paragraph contained a decent amount of technical jargon as well as common and widely used words. While the software was decoding the speech into text, we observed the CPU utilisation and also the total amount of time the software took initially to train itself.

After the first time that we read out the narrative, the technical jargon which the software couldn't recognise was added to the vocabulary of the software. Now, the software was again tested with the same narrative to see whether it could now detect all of these words. We used the same narrative a third time around to eliminate problems due to improper narration and measured the accuracy of the program.