| Helpful hints for developers using alphanumeric capture in a Form element |
| Categories: Audium Studio (3.4.x) |
| Article ID: | 81 |
| Last updated: | February 28, 2006 |
| User Opinions |
|
No users have voted.
|
|
Thank you for rating this answer.
|
SUMMARY
This article includes suggestions that may be helpful to developers who are creating Audium applications that capture alphanumeric input (e.g. "4UD1UM") from callers.
SYMPTOMS
Developer is creating an Audium application that includes a Form, and needs to capture alphanumeric input with it.
RESOLUTION
The technology for alphanumeric recognition has been available for a while, but you should use extra caution in making decisions on how to go about implementing it. Alphanumeric recognition is known as one of the most challenging tasks over the phone, due to the large number of similar-sounding alphabetic letters. For example, distinguishing the letter "t" from "c" or the letter "b" from "d" can be very difficult, even for human ears.
To tackle this challenge, a popular recommendation is to use military spelling, or use US state/city names which also include a finite set of options but tend to be more familiar to common application users. For instance, you can instruct the caller to spell with examples such as "A as in Alpha" or "A as in Alabama". After the input is collected, you can use n-best to confirm the choice with the caller when the recognition returns many possibilities. As much as possible you should try to restrict the letter and digit combinations in your grammar, based on either the type of strings you are looking for or information such as user account numbers and passwords that are stored in your database.
One alternative is to use Scansoft's Alphanum OpenSpeech Dialog Module (OSDM). This OSDM is designed to recognize a connected string of up to 20 letters and digits, where the sound "oh" is recognized as a synonym for both the digit zero and the letter "O". To help improve recognition accuracy, the OSDM provides an "entries" parameter that allows you to constrain valid results by limiting recognition to a list of user-supplied strings. If at all possible, we would highly recommend that you take advantage of this parameter and restrict the recognition context. Because if the "entries" list is not specified, your recognition context will include all characters and numbers. This will almost infinitely expand the grammar coverage, but severely compromise recognition accuracy. Also, regardless of how the OSDM is configured, it is important to run a good amount of testing to get a realistic idea about the recognition performance and if possible tune and improve recognition accuracy. False recognition and false rejection rates are expected to be higher than other OSDMs. For more information about this OSDM (including how to obtain it) please contact Scansoft directly.
|
| Visitor Comments |
|
No visitor comments posted. Post a comment
|
| Attachments |
|
No attachments were found.
|
Search the Knowledgebase
Jump to ID
Browse by Category
Print this page
Post a comment