I have been getting latency problems in Dragon NaturallySpeaking version 11 even though I am using a very fast laptop (their XPS 1645 with Core i7 processor 1.6 GHz and 8GB of RAM. My microphone is a Bluetooth microphone (Sennheiser BW 900). I have used this combination with Dragon NaturallySpeaking version 10.1 without problems and was reasonably happy with it without any latency problems. I have also used the check links utility to remove any broken shortcuts. A copy of my Dragon Log is attached. I have even removed my antivirus software and removed any anti-spyware programs from my computer. Still, the problem persist Thanks




-------------------------

Dell Inspiron 7520 SE,  2ghz core i7(Ivy bridge), 8gb ram, 750gb Seagate Momentus XT SSD/HD Hybrid hard disk. SpeechWare 3 in 1, Sennhesser ME3, Buddy USB 6G, Windows 7 Ultimate, Office 2007.

supee,

There are several possibilities, but my first question is when you say "latency problem" your description is too vague. When you experience latency problems, what type of latency problems, and where?

First, I notice in your Dragon log that you converted a previous version user profile to version 11. The conversion process does not convert everything. In fact, the base Acoustic Model, vocabulary, Language Model and background dictionary (vocabulary) are completely replaced with the new version 11 speech models. Since there are significant changes to the way DNS 11 works with all of these speech models, there are certain things that you need to do relative to any conversion if you want to continue to use that converted user profile.

1.  Edit your custom words. DNS 11 no longer accepts hyphens or symbols in Spoken forms. If you have custom words that use hyphens and symbols in their Spoken forms, you need to remove these hyphens and symbols. Although these custom words may continue to function, the use of hyphens and symbols in Spoken forms is no longer permitted in DNS 11. There are situations and circumstances in which such custom words may not be recognized correctly. Also, the in-line (structured) commands; such as "new line", "all caps", etc. are now all lowercase with no hyphens.

2.  All built-in DNS commands are now lowercase. If you have created any custom commands that conflict with any of the built-in DNS commands, you may run into conflicts. Some of the DNS commands that were previously legacy dvc scripts are now Natural Language Commands. Therefore, you should edit your custom commands, if only for the purpose of making sure that any such custom commands that use previous version in-line in-line (structured) commands, such as HeardWord "\Cap" need to be changed to HeardWord "\cap" (uppercase to lowercase). Otherwise, such commands will not work in DNS 11.

3.  Any training you have done with your previous version user profile in that previous version, is not inherited when converting your previous version user profile to DNS 11 (i.e., general training). If you trained your previous version user profile, then you need to retrain it in DNS 11. This will bring it up to consistency with the new Acoustic Model algorithms. This can be a source of latency issues.

4.  In conjunction with #3 above you can experience latency issues relative to your microphone/soundcard. In your Dragon log there are a significant number of instances where the Audio Quality Monitor feature in DNS 11 is resulting in the rejection of your dictation because of low speech to noise ratios. When this happens, DNS can take a long time to resolve utterances and you can end up with latency issues where the circling cursor around the green flame continues to spin and you get a number of "Please say that again" responses, which show up in the Recognition History as <???> Entries. Be sure that you're using a microphone designed for speech recognition. The Sennheiser BW900 is, Akono HBH-300 is not. Regardless, if the audio quality of your microphone/soundcard is not optimal, this will cause latency issues. This is well-known with regard to pairing low quality, non-noise canceling, not designed for speech recognition microphones with DNS 11.

My recommendation is as follows:

1.  Create a new user profile without importing any custom words or custom commands and skip training. Simply test this new user profile for latency issues compared to your original converted previous version user profile. You find that the newly created user profile is less prone to latency issues then your original previous version converted user profile, then you have your answer. It is not unusual for previous version user profiles that are converted to version 11 also inherit any subtle or otherwise previous version user profile corruption or inherent bugs, flaws, or other negatively impacting issues. The conversion process does not eliminate these. In fact, it perpetuates them. This is the prime reason why we do not recommend using a previous version user profile that has been converted to DNS 11 – (v11) after your user profile name – because the conversion process does not fix any inherent problems in your previous version user profile, and yes, you can inherit these problems even if you didn't seem to experience them when using that profile in your previous version of DNS. Some of these are subtle, some of them are significant. Nevertheless, the conversion process does not fix them. They are drag right along with and into DNS 11.

2.  If you find that latency issues are reduced with a new user profile, export your custom words (vocabulary) after editing them as noted above and import them into your new user profile. It is best not to use Manage Vocabularies… to do this because all you will do is end up inheriting any inherent problems in your converted user vocabulary. Yes, this means that you may have to reset properties etc., but if there are inherent problems in the previous version converted user profile relative to the vocabulary, you will inherit them by using Manage Vocabularies to export and import such. Better to use the Export custom word and phrase list… and Import list of words or phrases… So that you avoid any possible re-corruption of your custom vocabulary. Also, export your custom commands using the Export/Import feature in the Command Browser vs. copying the mycmds.dat from your converted user profile to your new user profile. As with your custom vocabulary, there are conditions under which copying the mycmds.dat from one user profile to another can also inherit inherent issues. While it is less likely with regard to custom commands and the mycmds.dat, humor me for the sake of consistency and avoiding possible issues, whether they exist or not. In other words, better to be safe than sorry, pardon the use of a trite cliché.

3.  If desirable or necessary, run the general training on your new user profile. However, I would try using it first to see whether or not the accuracy is suitable and whether or not you experience any latency issues after importing your custom vocabulary and custom commands.

This may seem like perhaps a lot of work, but given your laptop configuration, and without knowing what specific latency issues you're experiencing, my guess is that you will experience excessive latency under some conditions because all users of DNS 11, regardless of the performance and capabilities of their computer hardware, experience some degree of latency. However, in such cases it's generally very short (normally less than 5 to 10 seconds) and usually with regard to the oversensitivity of microphones using the new 22.05 kHz 16 bit Mono sampling rate under certain conditions, such as waking up the microphone or the impact of certain backgrounds, such as people speaking in the background. However, using Bluetooth microphones, because of their limited frequency response, are generally more prone to latency issues, which is why I do not support Bluetooth microphones with any of my clients. As far as I'm concerned, Bluetooth is not ready for prime time with regard to speech recognition regardless of anyone's experiences. Bluetooth microphones or more prone to problems than any other designed for speech recognition microphone, and when my clients need good accuracy and performance upfront, I don't have the time to support them if they're using Bluetooth microphones, and they can afford downtime when using such.

Lastly, these are just recommendations based on working with hundreds of clients using DNS 11. It's your choice as to whether or not you want to try them or not. However, in most cases I find these suggestions reduce latency problems to the absolute minimal until the SP1 patch for DNS 11 rectifies those inherent issues that remain.

Chuck Runquist
Technical Project Manager
VoiceTeach LLC
Home of VoicePower® Ultimate

The significant problems we face cannot be solved at the same level of thinking we were at when we created them. - Albert Einstein

-------------------------

Scanning through your log file, there are a couple coming up. Please let's consider the following line as an example:

Opening user file C:\ProgramData\Nuance\NaturallySpeaking11\Users\HBH300 (v11)\current\voice\uk_seasian_english_two_core\eng_11k_p1.usr

This indicates that you are using a profile that has been upgraded from version 10, as is indicated by (v11), which may be doable but it is not recommended. Furthermore, and even more problematic, if I am not entirely mistaken, the profile has not been set up properly as a Bluetooth user profile, but as standard microphone/headset profile using the 22,050 Hz sampling rate (eng_11k_p1.usr). When using a genuine Bluetooth profile, this should read _8k instead, if I am not entirely wrong, but I don't think I am.

All in all, Dragon must have a hard time starting up, launching the profile and getting everything ready to go, so the best thing for you to do is set up a new profile anyway, and making sure that this will be done properly, instead of  going on using the current profile. Next thing, you might as well transfer all your custom data (words and commands) into the new profile and go from there.

This will make all the difference for you.

Rüdiger

 

-------------------------

Well, it's past the point where we can make any changes in the code, but we can still make changes to the Easter Egg!

Rüdiger,

You bring up a good point. I've never given much thought to what 11k means in terms of Acoustic Models in DNS 11. However, you may be correct.

Regardless, 11k would represent the Nyquist frequency range not the sampling rate (11,000 kHz) otherwise it would be 22.05k or 22k. Under that condition, 11.025 kHz 16 bit Mono would be 5.5k (5.5125) and Bluetooth would be 4k, not 8k (8 kHz 16 bit Mono sampling rate).

I'll pose the question to development. In the meantime, perhaps someone who is using a Bluetooth microphone specifically set up as Bluetooth, if they would be so kind as to do so, could post their Dragon log so that we can check it. I can't do this because I refuse to use Bluetooth anymore and I don't recall ever noticing what the settings were in DNS 11.

Chuck Runquist
Technical Project Manager
VoiceTeach LLC
Home of VoicePower® Ultimate

"At times we shall simply have to admit that, one way or another, what we can neither explain nor understand certainly doesn't cease to exist because we cannot see how it does or why it should." - Dr. Mark Hyman

-------------------------

Chuck,

I have been through the same math also, however, these (11k vs. 8k) are the only options under the Data, AM folder for any given language which is where the raw material is kept for creating the user profiles, and if you look at what the help files say under "Speech Model", it all makes sense.

Rüdiger

 

-------------------------

Well, it's past the point where we can make any changes in the code, but we can still make changes to the Easter Egg!

Rüdiger,

While I see what you're driving at, under the AM folder there are two entries for 11k (C:\ProgramData\Nuance\NaturallySpeaking11\Data\eng\am\eng_11k_p1.usr & C:\ProgramData\Nuance\NaturallySpeaking11\Data\eng\am\eng_11k_p2.usr), there is only one for 8k (C:\ProgramData\Nuance\NaturallySpeaking11\Data\eng\am\eng_8k_p1.usr), which seems to indicate that these are only referencing BestMatch III. So, the question remains what do these designations actually mean.

Also, even in the DNS Help, the 8k references are only to BestMatch III options Re: Bluetooth.

I still say it would be useful for someone who has a Bluetooth user profile who is using BestMatch IV to post their Dragon log so that we can see what is actually displayed in the log under this condition. That would help clear up the confusion.

Chuck Runquist
Technical Project Manager
VoiceTeach LLC
Home of VoicePower® Ultimate

"What you are aware of you are in control of; what you are not aware of is in control of you." - Anthony de Mello

-------------------------

Thanks to Chucker and Rudiger, I have just come back from work and will be working through both your suggestions, and will report back. A quick question: how do I transfer my custom words and commands?

Scrach that ; found the answer from the Vocabulary menu

Again I am very grateful for your help

-------------------------

Dell Inspiron 7520 SE,  2ghz core i7(Ivy bridge), 8gb ram, 750gb Seagate Momentus XT SSD/HD Hybrid hard disk. SpeechWare 3 in 1, Sennhesser ME3, Buddy USB 6G, Windows 7 Ultimate, Office 2007.

Thanks for your assistance, I have used the check links before but I realise that I was not waiting long enough for the links to be checked before closing the utility. I found nine dead links which I have now removed. I have created a completely new user, and I have had to set by the accuracy slider to only 50% and the performance is still a relatively poor and the accuracy is  poor. I am really getting frustrated with using this software. This is coming from somebody who started using dragon from version 1, though I do not consider myself an expert.

I have been using the Sennheisser Bluetooth microphone, and was relatively satisfied with the using dragon version 10.1. When I listen to make dictation, it seems to come out very clear. Can the microphone fail for seconds and become ok again so quickly?

Using this new user which I have trained by reading a lot of paragraphs, and setting the accuracy slider at only 50%, I realise that I still have to wait a few seconds for words to appear on the screen can also if I have to use the correction dialogue box. I have also adapted the user to my writings style

I am attaching a new dragon log file; started from yesterday, please check from this line:

"Dragon Systems error log started Thursday, March 24, 2011 07:31:05"

Thanks again

-------------------------

Dell Inspiron 7520 SE,  2ghz core i7(Ivy bridge), 8gb ram, 750gb Seagate Momentus XT SSD/HD Hybrid hard disk. SpeechWare 3 in 1, Sennhesser ME3, Buddy USB 6G, Windows 7 Ultimate, Office 2007.

Your BW 900 Bluetooth microphone should work as well in Ver. 11 as it did in Ver. 10 but unfortunately, it won't work much better because of the 5000 Hz frequency limitation. The Ver. 11 cutoff frequency is 11,000 Hz and for this reason, we recommend against using a Bluetooth microphone unless absolutely necessary.

Since you are using the BW 900, would you mind creating a new standard microphone or USB (depending on which you are using) user profile and electing to skip general training so that you can immediately find out how well the new user does? You may also be having an issue with your computer's built-in Bluetooth.

A 1.6 GHz computer is a little slow but you should have decent speed running at 50%. You might want to consider creating a user profile based on Best Match III rather than Best Match IV on this system.

When NaturallySpeaking has a rough time understanding your dictation, it will typically slow down your computer, by as much as 4 times, so it will appear that you are experiencing latency issues when the problem is the sound system.

Last recommendation: Don't waste your time reading lots of paragraphs or going through continuous training. It not only doesn't help but can even hurt your accuracy. We never recommend more than 1 training session unless you are suffering from substandard accuracy and only then, do we recommend 1 additional training session. We have the advantage of having a rather boring mid Continental US accent but it has 1 advantage; the ability to skip general training and get up and running, with a highly accurate new user profile, in about 30 seconds. Ver. 11 really should be that good but we've never had very good luck using the BW 900 as a Bluetooth microphone. We recommend using it as a 1.9 GHz with a USB soundcard.

-------------------------

Click KB 2012 REV D to Download a 30 Day Evaluation of KnowBrainer 2012
Click SpeechStart+ to Download a 15 Day Evaluation of SpeechStart+

Lunis,

You're not looking at his computer hardware specs very carefully. He's running a Core™ i7 Q720. That is a Quad core Core™ i7 mobile with 6 MB of L3 cache, 8 MB of RAM and dual channel memory, which gives him the equivalent of eight cores (hyperthreading). 1.67 GHz is irrelevant in his case. He shouldn't have to make any exceptions with regard to settings. BestMatch IV and the Speed vs. Accuracy slider set all the way to 100% should work just fine on his system. You're looking at CPU speed, which is basically irrelevant with Dragon 11, or any version for that matter.

This problem is microphone, not processor.

Chuck Runquist
Technical Project Manager
VoiceTeach LLC
Home of VoicePower® Ultimate

If you hear the sound of hoofbeats, think horses not zebras.
variation on Occam's Razor (Law of Parsimony)

-------------------------

Point taken Chuck. We should've caught that. We are just about to end a 2nd 15 hour day in a row so we are probably getting a bit punchy.

-------------------------

Click KB 2012 REV D to Download a 30 Day Evaluation of KnowBrainer 2012
Click SpeechStart+ to Download a 15 Day Evaluation of SpeechStart+

Lunis,

No problem. Just didn't want to go any further before somebody made an unwarranted assumption.

Chuck Runquist
Technical Project Manager
VoiceTeach LLC
Home of VoicePower® Ultimate

"The least questioned assumptions are often the most questionable." -- Paul Broca

-------------------------