Speech Recognition eventually slows down, then turns itself off

bachware · 21 Apr 2013

I have Win 7 SP1 (not my desktop in the specs below, but an ASUS U50F laptop, x64 2.13GHz Intel Core i3 CPU, 4GB RAM). I'm deaf, so use it, after training it at home, to transcribe speech at church via the PA and Comtek, audio levels set not to go into the red. To begin with it works fine, but gradually slows down, with SR (sapisvr.exe) lagging farther and farther behind speech until the lag becomes 10' - 15', when the SR orb auto-switches to "Off". Sometimes I can jog it back on with the {Ctrl} + {Win} hotkey (have it set not to respond to "stop listening" audio), but after a while it refuses to respond to the start-listening key, and puts up a DB that says "Speech Recognition has encountered an error...", to shut it down and restart it, whereupon it again works for a short while, but eventually posts the same error after an even shorter usage time. Finally I have to coldstart, but the phenomenon eventually recurs in the same pattern: Simply logging off the profile and back on doesn't solve it.

It looks to me as if SR is using up resources, like a buffer and suffering memory starvation. I have it set to start at logon, so looked in the usual Registry place to find where sapisrv was being started, hoping to find a way of starting it in Realtime priority from there so it'd get favoritism in resource usage, but couldn't get it to work (can't reset priority after it starts manually via TM, which says "Access is denied"). Ideally there should be a technical advanced subsystem for SR that allows a user to tell the OS how much resources it can use, what priority it should get, etc. I'm a tech, but couldn't find any Registry settings for giving SR more RAM. Ideas/suggestions?

carwiz · 22 Apr 2013

I'm not a SR tech but have you checked to see what else is running on your laptop? Many OEM PCs have boatloads of "Bloatware" that start up and run in the background. Some AV programs can also be a burden on the system. Microsoft Security Essentials (MSE) is fast and integrates well with Windows (I wouldn't expect less). Another option is to increase the size of the pagefile and/or add more memory.

Faladu · 24 Apr 2013

I suggest a portable digital voice recorder (like the ones reporters use) that saves the recording as a file and has a USB connector to later transfer the file to a PC then run software to convert it to text, rather than do this 'ON THE FLY'.

I am no expert on sound recordings to text, but I assume the software is basically just doing that, anyway, and ON THE FLY recording/conversion I could see causing these issues.

This doesn't help if you wanted it RIGHT NOW of course.

I have a recorder made by Olympus that saves it as a file with USB, was about $50. and uses 1 AAA.