KnowBrainer Speech Recognition
Decrease font size
Increase font size
Topic Title: Digital recorder that I can attach to my headset?
Topic Summary: So that I can play back what I really said when Dragon has problems?
Created On: 05/22/2020 06:32 PM
Status: Post and Reply
Linear : Threading : Single : Branch
 Digital recorder that I can attach to my headset?   - Ag - 05/22/2020 06:32 PM  
 Digital recorder that I can attach to my headset?   - Lunis Orcutt - 05/23/2020 12:35 AM  
 Digital recorder that I can attach to my headset?   - Ag - 05/23/2020 10:40 PM  
 Digital recorder that I can attach to my headset?   - Ag - 05/24/2020 07:24 PM  
 Digital recorder that I can attach to my headset?   - R. Wilke - 05/24/2020 05:13 AM  
 Digital recorder that I can attach to my headset?   - Ag - 05/24/2020 06:47 PM  
 Digital recorder that I can attach to my headset?   - mobleguy - 03/03/2021 10:37 AM  
 Digital recorder that I can attach to my headset?   - Mav - 03/04/2021 01:53 AM  
 Digital recorder that I can attach to my headset?   - PG LTU - 03/04/2021 11:42 AM  
Keyword
 05/22/2020 06:32 PM
User is offline View Users Profile Print this message

Author Icon
Ag
Top-Tier Member

Posts: 647
Joined: 07/08/2019

The title and topic summary say pretty much what I want:

 

In my  quesst to make Dragon more reliable, I would really like to have a separate authoritative recording of what I said when Dragon  and related tools start misbehaving.

 

recoognition history  and " play that back"  help a lot, frequently I have said things that I am pretty sure  are missing from the playback (e.g. in DragonPad) and which did not get recorded in the recognition history.

 

Basically, I suspect that stuff running on the PC  may be interfering with the reception of  wireless/audio/.../recognition/...

and I expect others will chime in saying " it's not the PC or  its software, it's RF interference".

 so be it.   in either case, something has happened to what I have  said between my lips and Dragon and the target application or operating system commands. Relying on audio recordings made on the PC may not accurately represent what was said.

 

Hence, I am looking for some sort of digital recorder that I can attach to my headset,  or which I can wear at the same time as I am speaking into my headset for Dragon it.    e.g. a lapel mic,  probably running a wire down into my pocket.

 

 Probably not  transmittiing wirelessly, certainly not to the PC on wwhich I'm running Dragon.   probably recording locally.   although extra credit iff it can record locally and also transmit to some other computing device to make it easy to get the recording and compare it when there is a problem.  That path can fail, but it would be convenient  for a quicker debug cycle.  when it fails, fall back to the local recording.

 

 ideally, I couldd take tthee recording made to this device, and play it back into my headset mic,  to see if I get different recognition the next time around.

 

 

My hypothesis being, if recognition is different the nextt time around, but the window context etc. is the same, then misrecognitions are being caused by some sort of dynamic workload charaacteristic. Like interrupts getting lost.



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

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.



 05/23/2020 12:35 AM
User is offline View Users Profile Print this message

Author Icon
Lunis Orcutt
Top-Tier Member

Posts: 39353
Joined: 10/01/2006

As far as making a recording is concerned, you can toss nearly any digital recorder into your shirt pocket, hang it on a neck chain or add an inexpensive TP 7 lapel microphone. You can optionally extend your microphone range by switching to conference mode. HOWEVER, in our opinion, this is a wa$te of money and time.


Have you considered opening your Dragon Options/Data tab and changing Save recorded dictation with document from Never back to Always or Ask Me. Then dictate, to your hearts content, in Microsoft Word or DragonPad. Dragon will embed a recording of your voice with your document text so you can play back immediately or later by selecting the text you wish to hear (in your own voice) by saying play back

 



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

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



 05/23/2020 10:40 PM
User is offline View Users Profile Print this message

Author Icon
Ag
Top-Tier Member

Posts: 647
Joined: 07/08/2019

Originally posted by: Lunis Orcutt As far as making a recording is concerned, you can toss nearly any digital recorder into your shirt pocket, hang it on a neck chain or add an inexpensive TP 7 lapel microphone. You can optionally extend your microphone range by switching to conference mode. HOWEVER, in our opinion, this is a wa$te of money and time.

 

 

Have you considered opening your Dragon Options/Data tab and changing Save recorded dictation with document from Never back to Always or Ask Me. Then dictate, to your hearts content, in Microsoft Word or DragonPad. Dragon will embed a recording of your voice with your document text so you can play back immediately or later by selecting the text you wish to hear (in your own voice) by saying play back

Yes, I have been doing that.

 

However, much of the time I am not dictating into a single document. Much of the time I'm bouncing between multiple  documents/programs/...   also,  many of my biggest frustrations with Dragon relate to commands.  as far as I can tell, the commands are not recorded in the saved dictation in the document.

 

What I am looking for is a source of truth completely independent of the PC. 

 

E.g. a time stamped recording of what  I said and when I said it.   but I could, for example, compared to  dragon's timestamp recognition history. So that I can quantize the annoying pauses by taking  the difference between when I said something in the time recorded in the recognition history.  and possibly correlate to system logs.

 

Even without the timestamps, I would like to have an accurate recording of what I said. I do not trust dragon's recording  of what it thinks I said, or what it received. I have seen (or, rather, heard)  far too many instances of  gaps and dropouts in such recordings.

 



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

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.



 05/24/2020 07:24 PM
User is offline View Users Profile Print this message

Author Icon
Ag
Top-Tier Member

Posts: 647
Joined: 07/08/2019

Returning to Lunis's suggestion  of using Dragon's embedded recording...

 

If this were a newsgroup thread I would give it a new/subject line.  i.e. try to create a new thread.  but I'll leave it here.

 

Have you considered opening your Dragon Options/Data tab and changing Save recorded dictation with document from Never back to Always or Ask Me. Then dictate, to your hearts content, in Microsoft Word or DragonPad. Dragon will embed a recording of your voice with your document text so you can play back immediately or later by selecting the text you wish to hear (in your own voice) by saying play back

 

while this has the obvious disadvantage of not being a source of truth independent of the PC workload, it has some advantages, especially convenience,  but also for usage models  other than  investigating accuracy problems.

 

I have been thinking about trying to do all of my dictation into Word or Notepad, and then transferring it  to whatever  nonspeech ready application I am working in,  typically OneNote or Thunderbird or  emacs or ... Because I'm lazy, I would probably want to write code  or at least  commands to do the transfer.   

 

Yes, in some ways this is just repeating the work of dictation boxes like DragonCapture. Moreover,  if command based I would lose the AutoPaste  feature of DragonCapture,  which is one of my favorite things.

 

However,  I would like to do something that DragonCapture does not do.  as far as I know DragonCapture and other dictation boxes, once you have pasted the text in the target application, forget about what you dictaated. I would like to remember what was dictated, tracking  a bookmark, the last place from which a transfer was done, and only transferring stuff  from the bookmark to the end of the file/buffer.

 

I would like that  so that, when I see that there was  a problem in the dictation that was transferred/pasted, at least I could go back and retrain it  in the Word  pseudo-dictation box.  Having the benefit of the recorded sound for playback.    It is not quite the same as  in a speech ready at, where you can correct and train at the same time.   But at least it would allow convenient training.

 

Like I said,  if I had to say a command  to do this transfer  it would  lose the benefit of DragonCapture's AutoPaste, which I quite like.  

 

It differs from the dictate into the box and then transfer model, the non-AutoPaste model,  and that the dictation has  already been transferred to the target application. It may be just me, but I  find it easier to correct things after the fact rather than before they are inserted the target application.    I frequently find it easier to do things after the fact rather than before or as an intermediate step.

 

I can imagiine having background code do the transfer, without requiring a. Last making it look more and more like a dictation box. In some ways, I'm just wanting a dictation box that doesn't forget as soon  dictation has been transferred,  which allows me to go back  and investigate misrecognitions in this dictation box history.  while it might not enable corrections to be propagated to the target nonspeech ready application, it might allow training, and vocabulary editing, e.g. creating helper custom  vocabulary to  disambiguate.

 

( I think it's quitte  likely  that somebody will  tell me " dummy, did you not know that the bar feature of the bar dictation box does what you askeed for?")

 

---

 

 however, even though it might be nice to have such a non-immediately forgetful dictation box,  allowing to go back to train or consider vocabulary editing for  errors, it would only help for stuff that Dragon considered to be dictation.   It would not help when Dragon has misrecognized dictation as commands (nor when the user has accidentally said a command while dictating).    for me, since I prefix most commands by "PUFF"  I don't have anywheere near  as  many  problems accidentally saying commands while dictating as I did  when I first started using  KnowBrainer.   my most common error now is for a command to be misrecognized, either as dictation, or, worse, as  a different command, possibly with dictation at the end.

 

This, however, is a "train/edit vocabulary while using" usage model.

---

 

 as for investigating  accuracy problems

 

(1)  I would like to have a recording BEFORE the mic

 

(2)  as well as a recording  of the sound before the  recognition engine,  i.e. just before  the stuff that gets put into recognition history, before  commands and dictation have been separated

 

(3) whereas  what Dragon associates with the document separation between commands and dictation

 

 three places in the pipeline.

 

 

 I doubt that this is available, but I can only hope or wish

 

 

 

 

 

 

 

 



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

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.

 05/24/2020 05:13 AM
User is offline View Users Profile Print this message

Author Icon
R. Wilke
Top-Tier Member

Posts: 7702
Joined: 03/04/2007

Even without the timestamps, I would like to have an accurate recording of what I said. I do not trust dragon's recording of what it thinks I said, or what it received. I have seen (or, rather, heard) far too many instances of gaps and dropouts in such recordings.


That much is true. Whenever what you said sounds more like "noise" than speech to Dragon, that part will, for good reasons, be discarded in the resulting audio collected and maintained by Dragon, underlying and attached to the results object, which is also why you won't hear it when doing a "play that back" on-the-fly, or when listening to the audio part in a DRA file after it has been finished by the time you save a document (if the option is turned on), not mentioning how cumbersome and difficult to handle that would be.

If you don't trust me, here is a simple way to check this out. Start a digital recorder and leave it running for, let's say, 10 minutes, but only dictate into it randomly. Transcribe the audio in Dragon. Open the DRA file and listen to the audio. Optionally extract the audio part in WAV format, if you have a tool to do this. (I have.) You will find that all the sequences when you were not speaking are no longer part of the resulting audio, and the same is true for anything else which you may have considered "speech", but Dragon did not think so.

So basically no chance to do that from within or in combination with Dragon.

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



No need to buy if all you want to do is try ...

DragonCapture KB Download (Latest)
DragonCapture Homepage

 05/24/2020 06:47 PM
User is offline View Users Profile Print this message

Author Icon
Ag
Top-Tier Member

Posts: 647
Joined: 07/08/2019

Originally posted by: R. Wilke  So basically no chance to do that from within or in combination with Dragon.

 

That's what I figured. So if I am  interested in when Dragon noise  suppression  is suppressing sound that I think should be recognized as speech,  I would have to have a separate recording.

 

I would then take that recording, and play  it back in different situations  into my PC and Dragon.  If the system workload  affects recognition ( or the reception of the sound pre-recognition)  then the output will vary

 

Unfortunately, I can't ever expect what is being played back to exactly replicate my voice,  nor can I exactly replicate the workload running on the machine when  the recording from real life was made.  (Intel has technology to do this, but even that is only approximate.) But  at least I can see how consistent the playback --> recognition is,  as workload varies.

 

 several prooblems:

 

1)  correlating -  I can wish for timestamps,

 

2)  many  standalone voice recordeers are themselves voice-activated, so are themselves  dropping "uninteresting" intervals,  whether quiet or  sound not recognized as voice

 

3)   ease of use:

 

 ideally I could have this running, and when I start noticing problems, I might just "instantly" transfer the last one to five minutes of audio for replay recognition experiments.

 

 Rather than wanting to record hours of audio, and then having to make a heavyweight effort to correlate, I would probably want the recording to be made in a circular buffer manner.  But then transfer quickly.

 

 I think it's unlikely that such a device exists, although could probably be programmed into a hackable (open source)  voice recorder.

 

 more likely a recorder that does not attempt to elide nonvoice sounds, just compress them losslessly or without too much loss, probably with a buffer of a few hours.  But then lots of work trying to use this.

 

 

 

 



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

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.

 03/03/2021 10:37 AM
User is offline View Users Profile Print this message

Author Icon
mobleguy
Power Member

Posts: 60
Joined: 05/13/2019

I realize this is an older thread.
I am not sure if this will help.  But I know it is possible to run a another recording application on the computer at the same time you are running dragon, and while using the same microphone source as dragon.  I will let others in the forum speak to wether or not it is wise or not to do so. 

If you are interested in a seperate audio record of what you have dictated, or you just want to test to see what the audio sounds like it should work.  After recording, you can either play it back or you can export it to a file and use dragon transcription.

Audacity is a very powerful, open source application that can do this.  After installing, you go to preferences and select the same source as you are using for dragon to record. (you can also select another source if you to further test).

Audacity allows you to record from thouse sources using a number of different interfaces ( (Windows WASAPI, MME, Direct Sound)

I realize this just scraps the surface of what you are looking for.

 



 03/04/2021 01:53 AM
User is offline View Users Profile Print this message

Author Icon
Mav
Top-Tier Member

Posts: 440
Joined: 10/02/2008

I also just noticed this thread and had a new idea:

The DgnMicBtn object has 2 methods WaveCaptureStart() and WaveCaptureStop() I'vve used in the past for troubleshooting audio problems on the customer's end.

When activated, Dragon writes everything the microphone records to a PCM WAV file.

By recording the timestamp you called WaveCaptureStart() you can calculate the timestamp for a given offset in the recorded file.

Needless to say that these wav files can become quite huge if you let Dragon log audio for extended periods of time.

And it's very easy to forget that everything the mike picks up is being recorded, so it might rise some privacy concerns.

 

hth,

mav

 03/04/2021 11:42 AM
User is offline View Users Profile Print this message

Author Icon
PG LTU
Top-Tier Member

Posts: 2181
Joined: 03/21/2007

+1 Mav - that is quite simple and should be quite effective . . .

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




PG





Remember folks, my comments and this forum are for entertainment value only, please, no wagering or other reliance on the contents herein.  I permit no commercial use of my ideas (whether expressions or embodiments) without my written consent.

Statistics
32173 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 596 guests browsing this forum, which makes a total of 597 users using this forum.

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