Voice-driven data entry for Windows

Question

Background

I often do surveys and quizzes which use mostly multiple choice answers. I can end up with hundreds of these papers, each with 20-40 answers which need to be entered into some system so I can do some analysis on the data. I was originally looking for a way to design the forms to allow automated scanning/conversion but that has turned out to be a (near) dead end.

The problem

It simply takes too long to enter the data. The constant looking at the form to see what to type and looking at the screen to verify it is being typed correctly consumes hours (and it is really, really, boring work as well as being hard on the eyes).

The ideal solution

I would love to have an app (I guess it could even be a website) where I could configure the possible answers (A1 B2 C1 D3 are typical answers) and then I could simply read out the answers.

I recently tried with a Microsoft Machine Learning sample website but when I read out "A1 B2 B1 C4 D2 C2" it would listen and output "Hey one. B2 be one. see for." As you can imagine, this was quite frustrating.

The most critical thing is that the data is accurate and clean. That is "A1 A one" would not be acceptable.

Is there any software which runs on Windows (again, it could be a website but I would prefer installed software because I might not have an internet connection all the time) which can handle this task or am I dreaming?

Answer 1

This isn't a direct answer to the question you posed. However, I notice that you mentioned attempts to use scanning.

About thirty years ago I was working for a company that did clinical trials. In the course of doing some studies they received thousands of forms containing dozens of responses to scales like this one. The patients were asked to mark their responses in pencil somewhere along the line. At some point I came along and was horrified to notice that the company was processing these reports using rulers to make the measurements.

We bought a digital tablet and I wrote adjunct software for our clinical trials system so that data entry personnel could touch the tablet's stylus at each end of the scale then on the patient's pencil mark. The software would arrange to make the necessary inference for entry into the database.

It seems likely to me that the same general approach would work for you. It might, I think, be faster to process forms in this way that to read them too.

Answer 2

Microsoft speech recognition allows you to specify grammar to look for. This way you restrict recognizer to look on other variants like "see for" and only look for letter and digit. This should greatly improve accuracy of the input. It should look something like:

<rule id="letter">
    <one-of>
      <item> A </item>
      <item> B </item>
      <item> C </item>
      <item> D </item>
    </one-of>
  </rule>
<rule id="digit">
    <one-of>
      <item> 1 </item>
      <item> 2 </item>
      <item> 3 </item>
      <item> 4 </item>
    </one-of>
  </rule>

This is going to be very simple application to write, you can even do it in Visual Basic and Excel, however, you need to do some small programming to implement it the way you need.

Answer 3

You can use the voice-recognition software Dragon NaturallySpeaking (Windows, not free) and define voice commands to handle your typical answers. Note that you need the professional version to define advanced voice commands.