The LumenVox Speech Engine can run comfortably on most modern PC hardware, so long as that hardware is x86 compatible (note that LumenVox supports 64-bit operating systems for both Windows and Linux as of version 10.0). Determining the computing resources needed for a given application requires knowledge of several different variables in order to size the server needed to run the Speech Engine.
First, you must have some idea as to the maximum number of simultaneous users: this is the number of people that will be using speech recognition at any given time. Given good server hardware, we generally estimate that a single server can host between 50 and 250 users at once.
The variables in determining the number of ports that can be put on a single machine are: the size and complexity of the grammar, the length of the audio, and how frequently users are speaking to the system.
For instance, in a traditional IVR, users listen to prompts more than they speak. A user may very well be speaking only once every 15 to 20 seconds. During the rest of the call, the users are listening to prompts, thinking about their decisions, or waiting for the system to respond.
The less frequently an average user speaks, the more users that can be safely hosted on a single machine. The chart below should give you an idea of the relationship between the average time between speech recognition and the total number of users that can be hosted on a single machine, depending on hardware:
As you can see, the number of ports that can be safely hosted on a single machine increases with the length between speech recognitions.
This chart reflects testing on a server-class Intel Quad Core machine and a desktop-class Core 2 Duo machine, with 4 and 2 GB of RAM, respectively. The test used an audio file that was 1 second long and a grammar with 500 words in it.
If you desire to run the Speech Engine on lower-powered hardware, it is likely LumenVox can support it. The Speech Engine can run on slower CPUs and with considerably less memory: however, the number of simultaneous users will decrease and the time to perform a speech recognition may increase.
The Speech Engine, since version 9.0, uses considerably less memory than older versions and may run on systems with as little as 512 MB of memory, but again the system may be slower and able to handle fewer users than under ideal circumstances.
Like all of these hardware requirements, memory use varies significantly depending on number of users, length of audio, and grammar size and complexity. The Speech Engine generally uses around 200 MB of memory, but may require up to 1 GB or more depending on the application needs.