 The magazine of the Melbourne PC User Group Free Speech 98 George Skarbek gskarbek@melbpc.org.au |
 |
Actually it is not free ($79) and not that good. There is a new voice recognition program on the market, the Phillip's Free Speech 98 program. I have had some experience with other speech to text programs, the last one being over one year ago and was disappointed with the results. However, this is the newest and just released, and according to Phillips, the best, and now can run on a faster machine than the other programs.
Unfortunately, on my system, it did not work well and in my opinion is not yet a viable product.
The computer it was tested on was a 300A Celeron with 64 MB of memory. The 300A is the one with the cache and, according to Intel has 90% of the performance of a 300 MHz Pentium II. I did not have Phillip's noise cancelling microphone but used a good quality microphone and the tests were performed in a quiet environment. The tests that Free Speech 98 conducts on the microphone during installation gave the mike OK, stating that good results would be possible. For comparison, running the acceptance test on a very cheap mike that came bundled with a $20 software program was given only a medium rating and stated that very good results would not be possible.
The program needs about 80 MB of disk space, which includes about 20 MB for about half-an-hour's voice recognition training. The training files can be deleted if further training will not be used.
Tests below, using Free Speech 98, were performed after about 20 minutes of training. (The minimum recommended training time is 15 minutes.)
Original text that was read out
On the download page, you will see three files. The first file is the upgrade to your program. The second file is the documentation for your program (if separate from the upgrade file), and the third file is the Adobe Acrobat Reader program used to read the program's documentation. Download each file to your hard drive.
First attempt
The above text was read at normal-to-fast reading speed. Below is the result.
I download the age you as the three files. That this is for the upgrades to program. The second is the documentation off with Greg [is separate from the upgrade file], the root file is the Adobe bank about treatment program used to read the program's documentation to unload the check file to you how did try.
Second attempt
The same text was read at a slightly slower speed.
On the download page, you will see three fights. The first column is above grade school playground. The second problem is the documentation for your program [if said that from the upgrade file], and the head while using adobe bankrupt agree to play a grand used to read the program's documentation can lead to each file to you are dry.
Third attempt
After an additional nine minutes of training, and read slowly and carefully:
On the download page, you will cease three fights. The First ard s the upgrades to your program. The second is the documentation for your program [if separate from the upgrade file], and the third file is to Adobe record that read a program used to read the program's documentation. Download each file to your hard drive.
Although I cannot guarantee that when I read the text again, I did not mispronounce, or miss out a word, I am sure that my variation was not as large as these texts show. Note that the punctuations marks must be read out as comma, period. The brackets are read out as open bracket and close bracket.
This program is designed for the North American market and my accent is not
American--this could be the reason why the recognition is poor. However, I do not intend to do more training
(to improve recognition), as even a slight improvement is still not close for any business use.
Reprinted from the March 1999 issue of PC Update, the magazine of Melbourne PC User Group,
Australia
