Step 1: Install languages
A language provides needed resources to perform recognition in that language and includes localized:
- Acoustic models
- Pronunciation models and lexicon
- Standard VoiceXML built-in grammar types like date and digits
Voice Platform comes with the North American English language (en-US) by default. Before you begin localizing your system, you need to download and install any additionally required languages on your Nuance Speech Server hosts. Nuance recommends installing all languages on a single Speech Server host for multilingual deployments.
Languages are available for download from Nuance Network at . Click the Languages link under the Nuance Recognizer product.
Memory considerations in a multilingual deployment
Multilingual deployments must make sure they have enough physical memory and per-process user address space. When the Speech Server loads more than one language, that host must have 4 GB or more of physical memory, especially if the host is running many processes.
It is also important that to make sure that each process does not run more than the per-process user address space allowed by the operating system. This is 4 GB on 64-bit operating systems.
Languages, voices, and grammars must be provisioned so as not to exceed the user space limit. Additionally, reserve about 500 MB of headroom to allow for non steady-state events such as dynamic grammars, acoustic adaptation, and call log cleanup.
To respect the user space limit of 4 GB (and keep 500 MB of headroom) in a deployment environment, Nuance recommends that a Speech Server load only up to two languages. If a host has more than 4 GB of physical memory (for example, 8 GB of physical memory), the Speech Server can load up to three languages. If you are running more three languages, you must provision appropriately. For example, a host running five languages and three recognition services needs about 12 GB of physical memory.
