Required hardware and memory
This section lists the processor, memory, and storage requirements for a Nuance speech system in development and deployment environments.
Note: Nuance recommends more powerful machines for large deployments with high port densities and heavy call volumes.
Note: Nuance recommends explicitly configuring the date and time of all host machines in the network. Having all hosts use a common date and time simplifies log collection, licensing, troubleshooting, and comparing logs.
Processors for speech software hosts
Suggested processor requirements for speech software hosts are:
- Minimum: Dual-core Intel Xeon 2.0 GHz or Dual-Core AMD Opteron 2216 2.4 GHz (this CPU does not support using Vocalizer).
- Better: Dual Quad-Core Intel Xeon E5410 2.33 GHz (this CPU does not support using Vocalizer).
- Required for Vocalizer: Single instruction, multiple data (SIMD) instruction set extension with streaming SIMD Extensions (SSE v4.2).
Processors for Krypton hosts
Suggested processor requirements for hosts running the Krypton recognition engine:
- Minimum: Dual Quad-Core Intel Xeon E5410 2.33 GHz
- Better: 16 core Intel Xeon E5 Family v4 (Broadwell)
Requirements for development environment
For demos, prototyping, and small deployments (not more than 10 ports), you can install all Nuance products, including Management Station, if used, on a single host. Suggested configuration requirements are:
- Memory: 16 GB
- Processor: Dual Quad-Core Intel Xeon E5410 2.33 GHz
- Storage: Varies according to the port density. The range is between 10 to 40 GB.
Requirements for deployment environments
If using Management Station, install Management Station and Nuance speech software on separate hosts.
Management Station hosts
Memory and storage requirements vary depending on the volume of data you plan to collect and store. Recommended configuration requirements are:
| Memory | Storage |
|---|---|
|
|
|
|
Speech software hosts
This table shows the minimum and recommended memory and storage requirements for speech software hosts performing moderate logging and single-language recognition. Any host running Nuance speech software must provision for storing call log files, utterances, and whole-call recordings (if enabled).
| Memory | Storage |
|---|---|
|
|
|
|
Krypton and NTpE hosts
Memory and storage guidelines for dedicated hosts running the Krypton recognition engine performing moderate logging and single-language recognition:
| Memory | Storage |
|---|---|
|
|
|
|
> mount -o remount,size=3G /dev/shm
Adding memory for multilingual deployments
Multilingual deployments must make sure they have enough physical memory and per-process user address space.
- Minimum physical memory for a Nuance speech software host with more than one language: 4 GB
- Per-process user space on a 64-bit Linux host running 32-bit applications: 4 GB
Hosts running the Krypton recognition engine use a single port-per-process architecture for efficient memory management. Memory provisioning for the hosts is dependent on the number of ports provisioned with a recommended 512 MB of memory per port.
When provisioning languages, voices, grammars, and data packs, you must ensure they do not exceed the user space limit. In addition, reserve about 500 MB of extra space to allow for events such as dynamic grammars, acoustic adaptation, and call log cleanup.
In deployment environments, Nuance recommends loading a maximum of two languages on a Nuance speech software host. Doing this helps to respect the user space limit of 4 GB with 500 MB of extra space. However, this recommendation is flexible depending on the hardware. For example:
- You can load a third language on a 64-bit Linux host with more than 4 GB of physical memory (for example, 8 GB).
- To run more than three languages, you must provision appropriately. For example, a host with five languages and three recognition service instances needs about 12 GB of physical memory (including shared-memory).
