In using your outbound notification demonstration, I've noticed an extensive time delay between when the phone is answered and the message begins. This concerns me since I don't want people hanging up before the message starts.
How do you control that delay?
Thanks
We've Moved! Please visit our new and improved forum over at our new portal: https://portal.plumvoice.com/hc/en-us/community/topics
why is there a delay at the beginning of Outbound call?
delay explanation for beginning of IVR outbound call demo
There is an intentional delay at the beginning of the IVR outbound call demonstration in order to do answering machine detection.
This can be disabled easily, but not from the point of view of the person who is listening to the demo.
It's simply a matter of disabling the answering machine detection algorithm.
We are currently working on the first revision of our outbound IVR hosting service, which will offer the choice of enabling this algorithm programatically, but for now it is left enabled by default.
(see below for a more precise description )
The algorithm our engineers developed is a fairly common solution to the answering machine/voice detection problem. Here is the scope of the problem:
The initial algorithms for call type classification consisted of:
1) checking for noise on the telephone line.
2) if the noise is high enough , we have reached an answering machine
otherwise, this is probably a human.
But nowadays what with line improvements and better telephony equipment, the line signal quality is no longer a good metric for judging between a person and a machine.
The new algorithm involves listening at the beginning of the IVR call for a specific amount of silence (typically 3-5 seconds, the amount of time a typical person takes to pick up the phone, put it to their ear and utter "hello"). There is also some other additional checking that is done involving a 2 phase pass over the input data to further increase likelyhood of correct classification of the call type.
As you might expect, there is no "silver bullet" to perfectly distinguish between an answering machine and a human. (everyone has experienced the inevitable embarrassment of talking to an answering machine or voicemail system before realizing it's a machine on the other end of the call
). Plum has done extensive testing using a variety of test subjects, as well as various devices such as voice mail service providers, fax machines, and answering machines and received fairly good results in the range of 75% to 85% accuracy across all test participants and devices.
Hope this helps!
The Plum Support team
This can be disabled easily, but not from the point of view of the person who is listening to the demo.
It's simply a matter of disabling the answering machine detection algorithm.
We are currently working on the first revision of our outbound IVR hosting service, which will offer the choice of enabling this algorithm programatically, but for now it is left enabled by default.
(see below for a more precise description )
The algorithm our engineers developed is a fairly common solution to the answering machine/voice detection problem. Here is the scope of the problem:
The initial algorithms for call type classification consisted of:
1) checking for noise on the telephone line.
2) if the noise is high enough , we have reached an answering machine
otherwise, this is probably a human.
But nowadays what with line improvements and better telephony equipment, the line signal quality is no longer a good metric for judging between a person and a machine.
The new algorithm involves listening at the beginning of the IVR call for a specific amount of silence (typically 3-5 seconds, the amount of time a typical person takes to pick up the phone, put it to their ear and utter "hello"). There is also some other additional checking that is done involving a 2 phase pass over the input data to further increase likelyhood of correct classification of the call type.
As you might expect, there is no "silver bullet" to perfectly distinguish between an answering machine and a human. (everyone has experienced the inevitable embarrassment of talking to an answering machine or voicemail system before realizing it's a machine on the other end of the call
). Plum has done extensive testing using a variety of test subjects, as well as various devices such as voice mail service providers, fax machines, and answering machines and received fairly good results in the range of 75% to 85% accuracy across all test participants and devices.Hope this helps!
The Plum Support team
Last edited by support on Thu Feb 25, 2010 4:38 pm, edited 3 times in total.
Solution?
Just out of curiousity, is there any estimated date for when this update will be complete, because as is, the delay is far too long to keep a caller on the line...in fact, I've spent a few hours trying to figure out why my app wasn't working, but the only issue has actually been me hanging up too soon! Thanks a lot!
IVR outbound no longer using that speech detection algorithm
The plum IVR outbound is no longer using that speech detection algorithm, which means the delay is gone. This also means there is no viable speech detection in place, because so many customers complained about the 3 second delay. You'll have to implement it in your IVR application if you want it.
sincerely,
Plum Support
sincerely,
Plum Support