Sangoma Lyra AMD setup

[miike]

Has anybody used Lyra yet? I would like to give it a try but the Sangoma installation instructions do not cover vicidial integration.

Will it work with Vicidial?

[williamconley]

If it'll work with Asterisk version 1.4, it'll work with Vicidial.

Of course, Vicidial has a 1.8 asterisk Beta out, but I've not used it yet. Probably a link to it would have helped get you a more robust answer

[miike]

Im surprised there hasnt been some discussion on this already, as Matt Florell knows about it as there is a press release out (cannot post link)

All info (except vici integration!) is on the sangoma sites - it looks ready to go. A single box vici with working AMD would be fantastic for us all.

[williamconley]

Is Lyra free?

[mflorell]

"Lyra" is just a new brand name for Sangoma CPD. It isn't new or even different in any way from the existing Sangoma CPD that we have been compatible with for the last 4 years.

It isn't free, it's commercial closed-source software that is licensed per-port. You can buy perpetual licenses or monthly licenses.

[williamconley]

Is it a new version, or just a renaming?

We've installed the linux version a few times, and a new install method would be great (the old one was ... a pain, LOL)

[mflorell]

It's just a new name, targeted to the Asterisk market. Nothing about Sangoma CPD is changing, in spite of our attempts to get them to add a few much-needed features over the last two years(Features that the new Lumenvox CPD already has apparently).

[miike]

I thought this was a brand new product that was going to enable us to integrate the Sangoma CPD with Vici more easily into a single box solution.

[mflorell]

Nope, it's just a marketing trick, and it worked, they got your attention

[williamconley]

That would likely be the result of vague "pretty" product descriptions that allow you to read what you want into them.

[prince.fer]

we would like to purchase few license of Lyra. Does any one have we are ready to pay license cost.

[williamconley]

I think I remember a post from Matt that the product is dead. Plus: Our last tests showed that later AMD in Asterisk was as accurate as Sangoma's product anyway. Add to that: It requires a completely separate server to run just that once piece of software and required all calls to pass through that server.
More importnatly: Any flavor of AMD (Sangoma or built-in to Asterisk or any other version) adds a minimum of two seconds "dead air" to the front of the call. In Vicidial, you ordinarily have about 1 second dead air (barely noticeable). But add 2 more seconds? Prospects are now all annoyed at this "blantent robodialer call". This is the ring of death with anything resembling a sale or any form of cold call. You depress your success numbers more than you would waste in having agents press "VM" or "A" for each answering machine call. On every call, every agent must now climb back up the trust mountain before beginning any other task. You'll dramatically increase your quantitiy of "fuck offs" before finding a friendly voice.

[dspaan]

Plus: Our last tests showed that later AMD in Asterisk was as accurate as Sangoma's product anyway.

How did you measure this and what was the % of accuracy? I remember vicidial group telling us that that apart from 'detecting' 80% of answering machines a significant part of that would be false positives, meaning humans who were wrongly flagged as answering machines.

[mflorell]

We never got over 80% AMD accuracy with the built-in Asterisk AMD, and that was with up to 5 seconds of analysis time, which lead to a huge hangup percentage of human-answered calls(often over 50%).

As for Sangoma's CPD(CPA/Lyra/etc...) even though you can't buy it anymore since Sangoma stopped selling it a while ago, it had very fast analysis time(often less than 1/4 second) and was up to 95% accurate. It also started analyzing the call before it was answered, which gave it a distinct advantage over Asterisk AMD. On the downside, it was temperamental, expensive, didn't work with all SIP carriers, and was limited to 200 channels per proxy server.

[dspaan]

Why did Sangoma discontinue this product and why are there no alternatives? You would think there is a market for this?

The AMD in vicidial do you set the analysis time somewhere or does it vary per call dynamically?

[williamconley]

Why did Sangoma discontinue this product and why are there no alternatives? You would think there is a market for this?

The AMD in vicidial do you set the analysis time somewhere or does it vary per call dynamically?

Our final tests for the "comes with Asterisk 1.8 and beyond" AMD was over 95% accuracy in the markets we tested (client-specification using their real-world data, live on their system). Most of the mistakes were "thought it was human instead of machine" rather than "thought this human was a machine" (which was actually quite rare).

Our tests involved using the built-in AMD with recording artificially turned on BEFORE the answer (to be sure we got all results) and then having a human listen to each result for verification. Several hundred calls. 95% accuracy.

In ALL cases the analysis time was 2 seconds. Anything below that was wrong so often it was useless (we didn't bother testing, it was that bad), and anything above that was not a notable improvement (never tested, though). While I admit the tests were all US numbers with major carriers so this is hardly "universal", the test results were enough to make the client in question drop the Sangoma idea quickly. Which was a shame since we were in fact going to get paid for the Sangoma Server buildout, but it never happened.

Over two years ago, if I recall correctly. Noone has asked us to run full tests since. Especially now that the Sangoma Product no longer exists.

I suspect they killed the program due to lack of participation. It would be possible there was litigation involved (ie: "screw it, we didn't make enough on this product to pay for laywers and coding this year!", I just doubt it, lol)

[mflorell]

For Sangoma I'm sure it wasn't worth keeping the support people trained on it for the amount of money they were bringing in. I did mention it was temperamental, it operated well within very specific parameters on only some carriers, and outside of those parameters/carriers it had all kinds of issues. Also, we sold very few licenses for it over the 8 years that we were actively selling it. If we had to maintain any staff on it ourselves, it wouldn't be worth the cost either.

As for other paid alternatives, we're working on that, but there are the same issues: not much demand at the price, and a learning curve to be able to support and maintain it.