Comments on: IVRS and Speech Recognition

By: Abhishek Mittal

Abhishek Mittal — Sat, 27 Mar 2010 11:35:05 +0000

Why we should speak digits as choices for IVR. Since we face then the problem of speech recognition. There are normally maximum 5 choices for a typical IVR system. I think that then we can use a same pronunciation for every caller but different duration of that voice.
Lets take an example:
We will use pronunciation of Hindi language’s “अ” character. It’s pronunciation is something like English language’s “A” character’s pronunciation.
Now my suggestion is depicted in the following table :
Choice Duration in seconds
1 less than 1
2 more than 1 and less than 2
3 more than 2 and less than 3
4 more than 3 and less than 4
5 more than 4
For choice 3, user will speak “अ” for more than 2 seconds and less than 3 seconds.
Means IVR system will use the time duration of spoken voice by customer.
Here, for choice 4, there can be slight problem to customer but upto 3 and for 5, i think there is no problem.
Can it be implemented ?

By: Uttam Pegu

Uttam Pegu — Mon, 16 Mar 2009 01:39:46 +0000

Hi Rob,
Thank you for the feedback. Accuracy of data seems to be very important in clinical trials. We have tried speech recognition on our few voice portals and the successful recognition never went above 80% even for regular callers.

By: Shailender

Shailender — Sun, 15 Mar 2009 16:41:32 +0000

It is really difficult for speech recognition technologies to be accurate in India where there are more than 250 languages! Many voice portals here gives out quite absurd result and the relentless confirmation from IVR is quite irritating.

By: Rob McGrady

Rob McGrady — Sun, 15 Mar 2009 03:02:58 +0000

IVRS in the clinical trial arena normally has callers in fixed locations (hospitals, clinics, doctor’s offices, etc) and (more importantly) requires that callers enter very specific data about the patients they are enrolling into a trial (from height and weight to age to health status). This data can be plugged into an alogorithm that determines the type and amount of medication the IVRS will assign to a patient. For these reasons, I have not seen voice recognition used in clinical IVR systems. If anything, users will prefer a web-based application (IWRS) in place of the phone. As you say, the reliability of existing voice recognition tools is not high enough to employ them in clinical trials. Good blog!!