KnowBrainer Speech Recognition
Decrease font size
Increase font size
Topic Title: Do I need to train custom phrases?
Topic Summary: Or is it sufficient to just add the custom phrase to the vocabulary?
Created On: 07/03/2021 12:32 PM
Status: Post and Reply
Linear : Threading : Single : Branch
 Do I need to train custom phrases?   - kkkwj - 07/03/2021 12:32 PM  
 Do I need to train custom phrases?   - Lunis Orcutt - 07/03/2021 01:05 PM  
 Do I need to train custom phrases?   - kkkwj - 07/03/2021 06:39 PM  
 Do I need to train custom phrases?   - Alan Cantor - 07/03/2021 09:28 PM  
 Do I need to train custom phrases?   - kkkwj - 07/03/2021 11:03 PM  
 Do I need to train custom phrases?   - Lunis Orcutt - 07/04/2021 12:57 PM  
 Do I need to train custom phrases?   - kkkwj - 07/04/2021 05:38 PM  
 Do I need to train custom phrases?   - Ag - 07/09/2021 04:26 PM  
 Do I need to train custom phrases?   - Ag - 07/09/2021 04:58 PM  
 Do I need to train custom phrases?   - kkkwj - 07/09/2021 06:47 PM  
Keyword
 07/03/2021 12:32 PM
User is offline View Users Profile Print this message

Author Icon
kkkwj
Top-Tier Member

Posts: 786
Joined: 11/05/2015

I have many custom phrases added to my vocabulary and I dislike spending 1/2 hour to train them whenever I create a new profile.

 

The reason I have custom phrases is because Dragon has a problem recognizing them because the word combinations do not follow the natural frequency of adjacent words in English language.

 

When I add the custom phrases to the vocabulary and train them, Dragon has no problem recognizing them. But now I am wondering if it is the existence of the custom phrase in the vocabulary that helps Dragon recognize them or the training of the pronunciation of the words that helps Dragon recognize them.

 

I am thinking that because the words are all common words that Dragon can easily recognize, it does not help Dragon very much when I train those phrases. 

 

Instead, I am guessing that defining the phrase into existence gives Dragon permission to recognize my phrase by overcoming or ignoring the natural word frequencies.

 

Does anyone know if it is the existence of the custom phrase that makes it work? Or are both the existence and the training required to make it work? Thank you 



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

Win10/x64, AMD Ryzen 7 3700X, 64GB RAM, Dragon 15.3, SP 6 PRO, SpeechStart, Office 365, KB 2017, Dragon Capture, Samson Meteor USB Desk Mic, Klim and JUKSTG earbuds with microphones, 3 BenQ 2560x1440 monitors, Microsoft Sculpt Keyboard and fat mouse

 07/03/2021 01:05 PM
User is offline View Users Profile Print this message

Author Icon
Lunis Orcutt
Top-Tier Member

Posts: 39191
Joined: 10/01/2006

We deplore training and only recommend it as a last resort. Even when training is your only option, Dragon occasionally has the memory of a goldfish. Vocabulary phrases are almost bulletproof without training and you can take it to the next Dragon release. Think of it is being able to take your Mercedes to heaven with you. You are wasting your time training. We have been using Dragon (NaturallySpeaking) since 1996; a year before it was released to the public. The 1st phrase we added was “USB Pod”. Attempting to train the word “pod” was useless. Dragon insisted on typing “pot” which will not be caught by your spell or grammar checker but will be caught by your mother-in-law who will likely remove you from the family will  We had a significant incentive to fix this issue. Dragon had no problem with the word “USB” so we added the phrase “USB Pod” to our vocabulary. Also note that the word “pod” is capitalized. We frequently use this phrase in place of “USB soundcard” which undoubtably would have been a smarter move but we are not typically accused of being intelligent at “KnowBrainer”. To quote the Big Haired Lady “company names can be more revealing than they seem”. Of course this is a woman who thinks fine wine comes in a box   The best part is that we have never had to re-create or correct that phrase. We just import our personal vocabulary into each new Dragon release. 



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

Change "No" to "Know" w/KnowBrainer 2020
Trial Downloads
Dragon/Sales@KnowBrainer.com 
(615) 884-4558 ex 1



 07/03/2021 06:39 PM
User is offline View Users Profile Print this message

Author Icon
kkkwj
Top-Tier Member

Posts: 786
Joined: 11/05/2015

Thanks, Lunis! I think you probably saved me a half-hour on the new profile cycle.

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

Win10/x64, AMD Ryzen 7 3700X, 64GB RAM, Dragon 15.3, SP 6 PRO, SpeechStart, Office 365, KB 2017, Dragon Capture, Samson Meteor USB Desk Mic, Klim and JUKSTG earbuds with microphones, 3 BenQ 2560x1440 monitors, Microsoft Sculpt Keyboard and fat mouse

 07/03/2021 09:28 PM
User is offline View Users Profile Print this message


Alan Cantor
Top-Tier Member

Posts: 4144
Joined: 12/08/2007

My experience has been that I rarely need to train phrases that I import into Dragon. They almost always work from the get-go.

When I'm specifying a spoken form, I find it helpful to use whatever word (or words) Dragon comes up with before I've added the custom phrase.
 07/03/2021 11:03 PM
User is offline View Users Profile Print this message

Author Icon
kkkwj
Top-Tier Member

Posts: 786
Joined: 11/05/2015

Ooooo, Dragon is much more snappy now! And I didn't even have to train all the words in my custom vocabulary. thank you Lunis for the encouragement! I have been avoiding this for a long time.

 

Alan, that's a great idea. I have only trained one custom phrase so far. My problem is that I have parallelism among many phrases that contain "h one" "h two" "h three", and so on. Dragon gets "h one" and "h three" right, but sees "H2" instead of "h two". To preserve the parallelism, I just trained the "h two" phrase and Dragon got it right immediately.

 

While I am here, discussing a new profile, I think I confused my old profile in the following way. I checked the option box that brings up the correct menu when I selected words. I don't think that was a smart move because I would select words to replace them with entirely different words. However, because I did that through the Correction box, I think maybe Dragon interpreted my changes as corrections rather than simple replacements. that might have screwed up my profile recognition over time.

 

I do not check that option box any more. I just select the words by voice and redictate over them.

 

 



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

Win10/x64, AMD Ryzen 7 3700X, 64GB RAM, Dragon 15.3, SP 6 PRO, SpeechStart, Office 365, KB 2017, Dragon Capture, Samson Meteor USB Desk Mic, Klim and JUKSTG earbuds with microphones, 3 BenQ 2560x1440 monitors, Microsoft Sculpt Keyboard and fat mouse



 07/04/2021 12:57 PM
User is offline View Users Profile Print this message

Author Icon
Lunis Orcutt
Top-Tier Member

Posts: 39191
Joined: 10/01/2006

If you have any doubts, just use Select-&-Say correction which is 3 times faster and Dragon understands that you are just editing. Note that you will need to remove the default checkmark from the Dragon Correction/Options Select commands bring up Correction menu and add a checkmark to Select commands search backwards. In our opinion, the Dragon default options include several bad choices.



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

Change "No" to "Know" w/KnowBrainer 2020
Trial Downloads
Dragon/Sales@KnowBrainer.com 
(615) 884-4558 ex 1

 07/04/2021 05:38 PM
User is offline View Users Profile Print this message

Author Icon
kkkwj
Top-Tier Member

Posts: 786
Joined: 11/05/2015

Thanks Lunis; your description matches my options exactly - I search backwards for selects, disable the Select brings up Correction box option, and just redictate over my non-optimal word choices.

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

Win10/x64, AMD Ryzen 7 3700X, 64GB RAM, Dragon 15.3, SP 6 PRO, SpeechStart, Office 365, KB 2017, Dragon Capture, Samson Meteor USB Desk Mic, Klim and JUKSTG earbuds with microphones, 3 BenQ 2560x1440 monitors, Microsoft Sculpt Keyboard and fat mouse

 07/09/2021 04:26 PM
User is offline View Users Profile Print this message

Author Icon
Ag
Top-Tier Member

Posts: 636
Joined: 07/08/2019

Piling on:

Lunis' advice to avoid training was one of the best things I picked up from this forum.

As was Alan's advice to use whatever Dragon came up with when specifying a spoken form.

Minor tweak: I do this when the misrecognition that Dragon came up with is uncommon[*] or something I would rarely say, e.g. Lunis' example "USB pod\\USB pot".

Doing this when the misrecognition is more common can cause problems. E.g. I use a lot of open source software, including the mail reader Gnus in GNU EMACS. When I say "Gnus" Dragon often comes up with "news" or "good news", depending on whether I pronounce the G and how much I emphasize it. I don't want a custom form "Gnus\\good news", since I occasionally use the phrase "good news".

So I have custom forms "Gnus\\guh news" and "Gnus\\guh nooz" [*] -- the "guh" for G requires more emphasis than I would naturally use, but is not too bad. I am willing to adapt my pronunciation somewhat to improve speech recognition.

Which brings me to my own piece of advice: I have had pretty good success creating "pseudo-phonetic" spoken forms, like "guh nooz". note that neither "guh" nor "nooz" are written forms recognized in my vocabulary. Dragon warns about this, saying "The vocabulary does not contain the following word. If the word is spelled as you intended, click OK and the program will give it an approximate pronunciation." this works well enough. there is no need to add "guh" as a custom word. Dragon does not added as a written form.

I find it useful to ask Dragon to "read that" back when I am trying to find a pseudo-phonetic spoken form.

--

As for acronyms: Lunis and Leslie advise using words like "bee" instead of "B" in the spoken form, but I have found exactly the opposite - better recognition when I use the letters. However, I find that it does not work to separate the letters in the spoken form "XYZZY\\X Y Z Z Y". I get better results when the letters are grouped e.g. "XYZZY\\XY ZZ Y". Typically pairs, although pre-existing forms like "XYZ" also work. I don't need to provide all possible groupings.

BTW, at first Dragon gave me "XYZ Z why", and I might very well have used that as the spoken form. But the grouped letter form works, and is easier to search for.

I nearly always find it better to have a naked acronym form as well as the broken up acronym form, e.g. "XYZZY" in addition to "XYZZY\\XXY ZZ Y"

--

Of course, I also consider compounds like "GNU EMACS" and "Gnus mailreader" - more context => better recognition - but I prefer to avoid this whenever possible, to avoid vocabulary bloat. Similarly "Gnus\\Gnus as in mail reader" and "Gnus\\Gnus with G". Sometimes useful, but custom vocabulary entries where the written form appears in the spoken form can cause problems.

--

Note *: BTW, I have "Gnus\\guh news" and "Gnus\\guh nooz", i.e. "nooz" as well as "news", because I seldom pronounce the words "news" as "nooz", but do pronounce "Gnus" that way before I started emphasizing the G to handle Dragon. My natural pronunciation of "news" is more like the Dutch "nieuw" followed by a Z sound. I have tried doing "Gnus\\nooz", but that causes problems for ordinary "news".

This is much like Dragon's insistence that "git" with a hard G be recognized as "get". AFAIK "git"-->"get" is an American regionalism. perhaps Dragon might not do this if I were using a British or Canadian vocabulary, but I really need American spellings with mostly Canadian pronunciation.


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

DPG15.6 (also DPI 15.3) + KB, Sennheiser MB Pro 1 UC ML, BTD 800 dongle, Windows 10 Pro, MS Surface Book 3, Intel Core i7-1065G7 CPU @ 1.3/1.5GHz (4 cores, 8 logical, GPU=NVIDIA Quadro RTX 3000 with Max-Q Design.

 07/09/2021 04:58 PM
User is offline View Users Profile Print this message

Author Icon
Ag
Top-Tier Member

Posts: 636
Joined: 07/08/2019

BTW, as I have long threatened, I have created a GitHub wiki to collect BKMs and speech recognition tools.

 

I have extracted my post above as wiki page https://github.com/AndyGlew/manx-UI--Speech/wiki/Training-Custom-Words-is-Usually-Not-Needed

 

This wiki is world readable,  but at the moment unfortunately  only I  can write to it.  I would like it to be community writable  if the usual issues can be resolved in a way that I am comfortable with.



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

DPG15.6 (also DPI 15.3) + KB, Sennheiser MB Pro 1 UC ML, BTD 800 dongle, Windows 10 Pro, MS Surface Book 3, Intel Core i7-1065G7 CPU @ 1.3/1.5GHz (4 cores, 8 logical, GPU=NVIDIA Quadro RTX 3000 with Max-Q Design.



 07/09/2021 06:47 PM
User is offline View Users Profile Print this message

Author Icon
kkkwj
Top-Tier Member

Posts: 786
Joined: 11/05/2015

Just as a follow-up to Lunis's advice not to train, I did not train my big vocabulary of custom words when I created a new profile a few days ago. Generally speaking, everything worked well. However, I had to train about a dozen custom phrases that Dragon misrecognized. I am keeping a list of words that I have to train when I create a new profile to make things more efficient.


But all in all, I am grateful for the Lunis advice that I did not have to train all my custom words.

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

Win10/x64, AMD Ryzen 7 3700X, 64GB RAM, Dragon 15.3, SP 6 PRO, SpeechStart, Office 365, KB 2017, Dragon Capture, Samson Meteor USB Desk Mic, Klim and JUKSTG earbuds with microphones, 3 BenQ 2560x1440 monitors, Microsoft Sculpt Keyboard and fat mouse

KnowBrainer Speech Recognition » Dragon Speech Recognition » Do I need to train custom phrases?

Statistics
32106 users are registered to the KnowBrainer Speech Recognition forum.
There are currently 1 users logged in.
The most users ever online was 12124 on 09/09/2020 at 04:59 AM.
There are currently 530 guests browsing this forum, which makes a total of 531 users using this forum.

FuseTalk Standard Edition v4.0 - © 1999-2021 FuseTalk™ Inc. All rights reserved.