Thursday, March 14, 2024

Review of "Live Transcribe and Notifications"

I have been experimenting with different speech recognition and dictation systems.  At one point I reviewed a Dragon dictation system.  More recently a friend recommended Otter.  I found Otter to be quite accurate, but the free version was too annoying to use for anything in particular.  It certainly wasn't useful for dictation.  I have used and reviewed Gboard.  It is not particularly accurate, and I definitely have to do an awful lot of editing once I have done the dictation.  But I have found it useful for doing dictation and the results have been worth the effort.

A limitation with both Otter and Gboard is that you have to be connected to the host company's servers, generally through the Internet.  Both systems have a simple front end interface, but all of the processing for speech recognition and transcribing is done at the back end.  That's where the heavy processing happens.  And so you need the connection. 

Now I have purchased a new phone, and it came with "Live Transcribe and Notifications," so I have a new system to try out.  And it promises that you can do dictation offline.  So I'm giving it a try.

In my first (offline) trial, I have found that it is not particularly smart, nor is it particularly accurate.  "And it has an annoying habit of Ending a sentence With a period. Whenever you Pause In the dictation, You have to keep speaking at a normal Rate And you cannot pause In order to get a An accurate result. The. Fragments I've sentences that it will take down And turn into text. May give you And idea of what you were talking about."  Actually, the more annoying part is that, even if it doesn't throw in a period, it will capitalize a word, as if you were *starting* a new sentence.  You have to do an awful lot of editing to "decapitalize" a lot of words in the text.

I will have to give it a try in terms of how much editing I need to do, and how much work is necessary to turn the transcribed material into useful text.  However, that shouldn't be a particularly unworkable problem, since you can, of course, just simply dictate multiple times or make additional statements, or say additional words in order to clarify what you were talking about if it gets anything wrong.  It does frequently make mistakes in taking down the words.  Often the words that you said will appear on the screen.  But, if you pause, it will decide that it was mistaken, and it may change the (correct) word to something else.  So having been an accurate transcription in the first place, it sometimes corrects itself and introduces a new error.  It's an interesting result. 

My little brother has recently asked about transcription services.  The enterprise that he works for is very concerned about security.  They have been using Zoom, and using the Zoom transcription service.  However, they have become concerned about the fact that the transcription service, operated by, and on, the Zoom platform, means that Zoom has access to the contents of their conversations and meetings.  (Personally, I think that Zoom's promotion of end-to-end encryption on the system is unlikely.  I strongly suspect that their encryption is only link encryption between Zoom and the individual users.  So I suspect that Zoom *does* have access to conversations and meetings held on Zoom, regardless of whether you use the transcription service or not.)  So they're looking at different options for transcribing, their meetings, and my little brother asked me for recommendations in this regard.  This particular program does seem suitable.  The offline transcription does Have some problems with accuracy.  But this transcription app does seem to be designed for meetings and for multiple speakers.  (And even background sounds.  That's what the "Notifications" part of the title is about.  It gives you notifications of sounds of birds, traffic, wind, and so forth.)  It is quite possible that this particular app could be something that they will want to use in this regard. 

No comments:

Post a Comment