Cepstral and app_swift behaviour

I have been working on a dynamic IVR system using Asterisk. I had designed the system on old hardware and plan on testing it on a production box. The system uses the Cepstral TTS engine with the app_swift module to repeat content back to the caller.

The old hardware has the following specs:
OS - CentOS 5
CPU - 900mHz p3
ram - 512 mb
Asterisk v1.4.17

The new hardware has these specs:
OS - RHES 4 (with HP Proliant drivers)
CPU - dual core Xeon
ram - 2 gigs
Asterisk v1.4.17

Now, the odd thing is that the old hardware runs the app_swift module fine. When I test it on the new, faster hardware, it seems to be very unresponsive and slow. One such example, when Allison speaks on the old hardware, the sentances are pretty fluid. On the new hardware, there is at least a 4 second delay between sentances. Also, if I hang up while she is talking on the old hardware, the call ends instantly (like it should). On the new hardware, there is a couple seconds delay, as if the app_swift module is trying to finish up, before the call is hung up.

Has anyone had any experience with this module? This seems like a very strange quirk. The only thing I can think of is the fact that the new hard is running an old version of GCC, whereas the old hardware has a more recent version of the compiler.

I think I resolved this issue.

After re-installing a couple times, I determined that the HP Proliant software was causing some kind of intererence. I only installed a few drivers, namely the NIC, RAID, and “multifunction” driver. I havent been able to pinpoint exactly what Proliant driver was causing the issue, but just having the bare essentials for an operational server seems to do the trick (I like running a tight ship anyways).

Also, I spoke with someone at Cepstral, and they informed me that using a concurrency license will also improve Swift performance because that license will allow the engine to handle more requests simultaneously.