![]() |
KnowBrainer Speech Recognition | ![]() |


|
Topic Title: CPU and RAM affects accuracy? Topic Summary: CPU and RAM affects accuracy? Created On: 06/18/2011 01:39 AM Status: Post and Reply |
|
![]() |
- bboyjkang | - 06/18/2011 01:39 AM |
![]() |
- Chucker | - 06/18/2011 06:47 AM |
![]() |
- Lunis Orcutt | - 06/18/2011 05:41 PM |
![]() |
- bboyjkang | - 06/19/2011 10:25 PM |
![]() |
- Chucker | - 06/20/2011 07:13 AM |
![]() |
- Alan Cantor | - 06/20/2011 09:28 PM |
![]() |
- Lunis Orcutt | - 06/21/2011 08:25 PM |
![]() |
- Chucker | - 06/22/2011 01:38 AM |
![]() |
- bboyjkang | - 06/22/2011 12:29 AM |
![]() |
- Chucker | - 06/22/2011 02:08 AM |
|
|
|
|
Can inadequate CPU and RAM affect Dragon accuracy? Or is accuracy mainly affected by your headset and the amount of accuracy training that you do?
Thanks |
|
|
|
|
|
|
|
|
Quote: Can inadequate CPU and RAM affect Dragon accuracy? Or is accuracy mainly affected by your headset and the amount of accuracy training that you do? bboyjkang, Generally speaking No. And inadequate CPU and RAM can affect performance (speed), particularly with regard to the time delay between dictation and display of text, but accuracy remains largely unaffected. By that I mean there are some instances where the time between dictation and display of text can affect your dictation style, which can cause you to dictate in short choppy phrases instead of using a natural style and dictating in longer more natural phrases, but accurately is mainly affected by how you dictate (dictation style), how clearly you enunciate and separate your words, and to a certain extent your microphone/headset quality and the quality of your Acoustic Model (training). Some users can skip training when creating a new user and get very adequate (optimal) accuracy without doing any general training whatsoever. Also, the last three versions of DNS have come with a more or less speaker independent base Acoustic Model, which simply doesn't require that the user do any additional training where the users dictation style and enunciation are smooth and clean as noted above, and the user has no accent or fails to use proper dictation style. Accuracy is a function of the speech engine (recognizer – MREC.dll). Performance has little impact on accuracy and what impact it does have is largely user based and not algorithmic [i.e., dependent upon performance (speed)]. Accuracy problems experienced by users with less than optimal CPU and RAM hardware configurations are generally a reflection of impatience, everything else being equal. Chuck Runquist Technical Project Manager VoiceTeach LLC Home of VoicePower® Ultimate ![]() "Many of the things you can count, don't count. Many of the things you can't count, really count." Albert Einstein -------------------------
|
|
|
|
|
|
|
|
|
Welcome to the KnowBrainer Forums
Just adding a smidgen to Chuck's sage advice… Because you haven't filled out your user profile, we don't know what version of DNS you are using or what you are using for microphone and soundcard. Although CPU and RAM have no effect, other than efficiency, on your accuracy, your soundcard & microphone are extremely important. The OEM microphone that's included with your software should be ONLY considered as a starter microphone and integrated soundcards often produce substandard results. It's easy to be fooled into thinking that additional training will help. Unfortunately, additional training often has the exact opposite effect. Besides proper dictation techniques, as noted by Chuck, knowing how to manipulate your personal vocabulary and optimizing your user file will have more effect. -------------------------
|
|
|
|
|
|
|
|
|
Thanks for the response. I have the Andrea Electronics NC-185VM USB Stereo PC Headset, and I've gotten some pretty wild results. I didn't do anything with the CD that came with the headset; I just plugged the headset right in. I also did the training for only 5 min. Yeah my processor is pretty bad, and I'm getting a new one pretty soon, but I thought that with the Andrea headset, the accuracy would be fine. Maybe it comes down to the training. I'm going to be using Dragon for command and control, and creating a custom vocabulary, so accuracy won't be too ultra important, but I'm still curious about it I passed the quality test, but I couldn't find the speech to signal ratio that I see people posting. Do you know where I can find this number? Thanks edit: not seeing my processor or headset at bottom but it's AMD Athlon 64 3700+ Processor (below minimum probably)
|
|
|
|
|
|
|
|
|
Quote: I passed the quality test, but I couldn't find the speech to signal ratio that I see people posting. Do you know where I can find this number? bboyjkang, You won't see the speech to noise ratio (Quality Check ratio number) at the end of the Check Microphone in DNS 11 Premium because Nuance removed this from DNS 11, and for good reason. That reason being that too many people rely on the quality check ratio rather than the overall Audio Setup Wizard process and looking at all of the other factors involved in microphone volume adjustment. In fact, some users were thinking that running the Quality Check actually adjusted your microphone even further without realizing that it's just a test and does not change anything. The measurement that you see reported, referred to, or contained in screen captures in other posts are from earlier versions of DNS (i.e., 10.1 and below). You will no longer see it in DNS 11 regardless of whether you're using Premium, Professional, Legal, or any other version of DNS 11. It's no longer there. Chuck Runquist "Many of the things you can count, don't count. Many of the things you can't count, really count." Albert Einstein -------------------------
|
|
|
|
|
|
|
|
|
Quote: In fact, some users were thinking that running the Quality Check actually adjusted your microphone even further without realizing that it's just a test and does not change anything. Running the audio setup wizard changes nothing? It doesn't even tweak the volume setting? |
|
|
|
|
|
|
|
|
Running the ASW/Microphone Check definitely resets the volume input level.
-------------------------
|
|
|
|
|
|
|
|
|
Quote: Running the audio setup wizard changes nothing? Tt doesn't even tweak the volume setting? Alan, I think you misread what I said. The Audio Setup Wizard (Check Microphone) consists of two basic parts: (1) the volume adjustment, and (2) the quality check test. The first part adjusts your microphone (volume) for best results with Dragon NaturallySpeaking. The second part checks whether that adjustment is adequate (quality). The second part does not readjust anything. It is simply a test of the volume adjustment and whether or not that meets the quality parameters. If the quality check fails, then in order to readjust your microphone (volume) you have to go back to the first part. Nowhere did I say that running the Audio Setup Wizard changes nothing, nor did I say anywhere that it doesn't adjust the volume. I simply said that the quality check (second part) is just a test. By itself that makes no changes. If the quality check fails then you have to go back to the volume adjustment in order to make any changes. This whole issue has been discussed so many times over the last 10 years that I don't even have a hard drive large enough to collect and save all the posts. So, it's important to read carefully what I said. Chuck Runquist Technical Project Manager VoiceTeach LLC Home of VoicePower®: Simply powerful, powerfully simple ![]() "Assumption is the mother of all screw ups" - Unknown -------------------------
|
|
|
|
|
|
|
|
|
Oh! I just noticed the speed versus accuracy slider under tools. I just started Dragon, I never touched the slider, but the program must have put it on fastest response automatically because of my poor processor. I guess in a way, a good CPU and amount of RAM can indirectly affect accuracy because you can crank that slider all the way to maximum accuracy.
|
|
|
|
|
|
|
|
|
Quote: Oh! I just noticed the speed versus accuracy slider under tools. I just started Dragon, I never touched the slider, but the program must have put it on fastest response automatically because of my poor processor. I guess in a way, a good CPU and amount of RAM can indirectly affect accuracy because you can crank that slider all the way to maximum accuracy. bboyjkang, First, the Speed vs. Accuracy slider determines how long DNS spends analyzing your dictation. Yes this can affect accuracy. However, even if the Speed vs. Accuracy slider set all the way to zero, if you dictate in longer phrases, sentences, or even paragraphs before pausing and use a natural smooth dictation style, then and even the fact that the Speed vs. Accuracy slider set to zero will still yield greater accuracy. It's a combination of factors. The reason that DNS initially sets the position of the Speed vs. Accuracy slider for CPUs and RAM that are less than optimal for DNS 11 is an attempt to improve performance (decrease latency). Regardless, even if the Speed vs. Accuracy slider is set all the way to 100%, dictating in short choppy phrases, single words, or dictating in an unnatural way won't improve your accuracy. Second, yes you can move the Speed vs. Accuracy slider all the way to 100%, but if your CPU and RAM are less than optimal, this will only increase the delay (latency) between dictation (speech) and transcription (display of text). In an of itself, setting the Speed vs. Accuracy slider all the way to 100% won't necessarily give you greater accuracy, but it will, in your case, allow you to go get a cup of coffee before the text is displayed. In cases like yours, the best setting for the Speed vs. Accuracy slider is 50%. On the other hand, learning to dictate in longer more natural phrases (i.e., in a style that mimics reading text as if you were reading an article from a newspaper or magazine) will provide better accuracy than simply setting the Speed vs. Accuracy slider. Basically, it's a trade-off. However, assuming that setting the Speed vs. Accuracy slider to 100% will necessarily give you more accuracy is a false assumption. Accuracy is more dependent upon your dictation style than anything else. -------------------------
|
|
|
|
|
FuseTalk Standard Edition v4.0 - © 1999-2013 FuseTalk™ Inc. All rights reserved.