DNS SDK Options

[GBNL]

Hi!

We have a software for which we would like to add voice recognition capability. I was thinking of getting the DNS SDK and start programming the interface. Since I have never really tried DNS before (only IBM ViaVoice which drove me away for voice recognition at full speed about 2 years ago), what kind of accuracy rate are you getting with the latest version of DNS on 1.6Ghz+ machines?

[John Wickett]

Processing speed helps, but RAM is more important. Count on 384-512 if on a network.

Most TRAINED users report 97-98% accuracy. New users are generally at 80-85% but it increases dramatically if a tool called Vocabulary Builder is run. This tool reads documents the user has created and analyses word/sentence structure etc.

John

[uclabloke]

What application are you making?

[GBNL]

Dental application written in VB6 for which the doctor has to input info in the database chairside. We would like to do away with mouse and keyboard with touchscreen and VR for the chaiside computers. Need to know if the technology (DNS's) is up there or if we are still dreaming...

[John Wickett]

You should enjoy NaturallySpeaking professional version 6. You use Visual Basic to create advanced commands.

I would suspect that the dentist would be using a wireless microphone.

John
Dictated with NaturallySpeaking---------------------------------------------jwickett@auracom.com
jcwickett@vac-acc.gc.ca

[uclabloke]

Awesome, good luck! Yeah, Dragon should be very close to 100% if you limit the vocabulary as much as possible.

Try deleting all the words you know the client may not need and that will reduce the chance of recognition errors. If your client has a very limited vocabulary, they will likely see 100% accuracy results.

As for hardware, I would recommend the AMD Athlon XP 1900+ (1.6ghz) processor. It has the best FPU of any chip in the market, which VR depends on greatly for performance. It also supports SSE. Check out the benchmarks here: http://www.anandtech.com/cpu/showdoc.html?i=1554&p=4


As for the motherboard I would go with MSI K7N420 (NForce 420 chipset) it has the best stability and memory bandwidth performance of any board on the market. Since it uses a dual channel memory system, Dimm 1 and Dimm 3 needs to be the place where you stick in the RAM. Dimm 2 [the middle dimm] needs to be empty for the best performance. I would go with (2 X 256mb) 512mb of Micron or Crucial PC2100 266mhz DDR RAM. http://www.pricewatch.com or http://www.crucial.com As for sound, Sound Blaster Live 5.1 or SB Audigy should get the job done.

I would also go with an Antec 300w Power supply, as they are widely known as one of the quietest and the most stable power source for the CPU.

For a list of the best priced parts vendors go to http://www.pricewatch.com/

Or if you would like to have it built for you MicronPC offers NForce Athlon XP systems http://www.micronpc.com/home/store/desktop...+_creative.html

It comes with integrated audio, so you may have to add SB Live PCI sound card.

However, http://www.compaq.com has PCs that you can configure to include PCI soundcards like SB Live with the Athlon XP systems.

[This message has been edited by uclabloke (edited 01-05-2002).]

[This message has been edited by uclabloke (edited 01-05-2002).]

[GBNL]

As for wireless microphone, yes it would be much preferable, any recommandation about the best brand/model anyone has tried?

[uclabloke]

Try this link out. They have a VR Wireless mike that seems up to the task.
http://www.speechcontrol.com/microphones/aws-100.asp

[John Wickett]

Rather than trying to delete words, which you couldn't do anyway because they are in a sense "hardcoded", just create a new vocabulary. You can load the specific words into the new vocabulary just by having the software read a particular document with the necessary vocabulary. If the program doesn't find the word in the new vocabulary, then it will look for it in the general vocabulary.

Unless you are a complete purist, don't worry too much about hardware. Just stay away from Celeron chips and integrated sound cards. I am quite happy with my Dell P. 4, 2 GHz processor and 1024 MB of RAM with a Sound Blaster live sound card.

John
Dictated with NaturallySpeaking---------------------------------------------
jwickett@auracom.com
jcwickett@vac-acc.gc.ca

[This message has been edited by John Wickett (edited 01-05-2002).]

[uclabloke]

Hmmm, maybe a open source speech engine where you can customize the vocab for your needs would be a possible alternative. But I don't have any knowledge on the matter.

Again the smaller the vocab, the better the recognition.