Working with data packs

Mix uses two types of data packs when building models for your applications:

  • Nuance data packs (NDP), used for automatic speech recognition and tokenization
  • QuickNLP (QNLP) data pack, used for semantic understanding

Nuance data pack (NDP)

A Nuance data pack (NDP), also known as a core data pack, is used for automatic speech recognition and tokenization.

Each data pack provides a base language model that enables the speech recognition engine and the text processing engine to recognize and transcribe the most common terms and constructs in the language and locale. The main identifier for this type of data pack is the Locale plus a Topic (gen for General in most cases). Additional topics are available upon request to provide a specialized, yet still general, knowledge of a domain or specific area of interest. Nuance data packs are used by both the ASR and the NLU engines.

When you add a new project in Mix you select the data pack topic and locale to use. If specialized topics are available to your organization, they will be listed for selection as well.

QuickNLP (QNLP) data pack

A QuickNLP (QNLP) data pack is used for semantic understanding. QNLP data packs enable the natural language engine to derive intent and reveal the meaning behind text and spoken input, by leveraging AI-based speech technology and powerful machine learning models.

QuickNLP data packs are used by the NLU engine. QNLP data packs contain predefined entities.

Predefined entities save you the trouble of defining entities that are common to many applications, such as monetary amounts, Boolean values, calendar items (dates, times, or both), cardinal and ordinal numbers.

Determining your application’s data pack versions

The data pack versions available to your application/project are listed on the Details tab in the Mix dashboard. From the Dashboard select your project and click the Details tab. The data pack versions are listed in the Information area.

When you create a project, each locale used in the project is associated with the current data pack version. When you rebuild your model, each locale continues to use the same data pack version. For example, if the version of the data pack is 4.7.0 for en-US and 4.1.0 for fr-CA when you create a new project, these data pack versions will continue to be used when you rebuild your model, even if the current data pack version available for a locale has increased.

Upgrading to a new NDP version

You may want to upgrade a data pack version to a new one or even revert to a previous version; for example, on the advice on your Nuance Mix representative. You can do this via the Mix dashboard. See Manage data packs.

Projects are associated with a specific version of a QuickNLP/NLU data pack. When you create a new project, it is initially associated with the current latest data pack version available for your engine pack version. Over time, the latest available QNLP data pack for your engine pack will update automatically without a need for any intervention on your part. When you rebuild your model within an existing project, the QNLP data pack for your project will update to make use of this data pack.

Data pack versions may not be fully backward-compatible. This means that if you create a new project but you want to import the TRSX content of a project associated with an older data pack version, you may encounter issues.

Upgrade steps for QNLP data packs, therefore, are not as straightforward as for NDPs. See Upgrade a QuickNLP data pack for more.