From: nospam@needed.invalid   
      
   On Sun, 11/16/2025 4:09 PM, pinnerite wrote:   
   > This is a speech to text program.   
   > Example:   
   > Dictated:  Fake Admiral at Remembrance Day Parade   
   > Printed:   They have more. Pat Remembrance.   
   >   
   > I couldn't see any feed back buttons for corrections like I used to have   
   with Dragon Dictate.   
   > I may give up.   
   >   
   > Alan   
   >   
      
   I got it working.   
      
   A balloon help from the OS, overlaying the "Settings" item at the top of   
   the menu, is why I missed the Settings at first.   
      
   I guess I was expecting to see an input selector on   
   the interface. As you can have more than one microphone   
   on a computer. And you may not want to be altering the   
   machine setting, just for one application cycle, then   
   setting it back. Once my microphone was selected in the system   
   audio settings, things improved after that.   
      
   Once I did that part, I got some results.   
      
      [Picture]   
      
       https://i.postimg.cc/LXrN3d48/Call-Me-Ishmael-Simple-Note.jpg   
      
   I set it for "continuous listening", but it still runs one sentence   
   at a time. There is a setting in the settings dialog that will   
   "sound a tone" when it is processing, and that would have allowed   
   me to read the first paragraph of the book "Moby Dick" into it.   
   But I didn't need to do that.   
      
   The phase   
      
       "If this is testing"   
      
   was repeated twice. On the first instance, the "If" got   
   clipped off. So to some extent it suffers from the   
   usual squelch effects. This could be solved by the software   
   winding back the input trace and starting conversion before   
   the first detect-able blurb occurs.   
      
   I could probably dictate into it, but it auto-punctuates, so I   
   doubt making an exact copy of Moby Dick is possible.   
      
   *******   
      
   Speech recognition was always a garbage-in garbage-out thing.   
   I tried years ago to do this, only to be thwarted by a microphone   
   that was for all practical purposes, as insensitive as a stone.   
   I have many duff microphones, I have one microphone used   
   for video conferences (a nuisance to setup), and I have a new one.   
      
   About a year ago, as an impulse purchase, I bought this big-assed   
   microphone. I brought it home, was not getting much signal, and   
   people were saying it was kind of a dud. One person successfully   
   described the tuning procedure (requires setting the gain a lot   
   higher than you might expect, which is why other people were dissing   
   it). I put it back in the box after tune-up. It's just been   
   sitting there.   
      
   This was the first opportunity to test it for real. And you   
   can see in my trace, that with the exception of the missing "If",   
   it is getting most of the input.   
      
   I used a heavy-weight model (from Whisper). The settings box DOES have   
   a hardware acceleration tick box... which is unticked, and requires the   
   user to tick it. The evidence is, it was working. Maybe. I don't   
   know if there is a graphical version of "nvidia-smi" I could use   
   to watch the video card utilization. My video card idles at   
   14W out of 180W.   
      
      Paul   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)