Real-Time Speech Analytics Model overview

A Real-Time Speech Analytics Model is a file that the Real-Time Speech Analytics Engine in the Real-Time Speech Analytics feature requires to detect specific words or phrases that are spoken in captured interactions.

A Real-Time Speech Analytics Model contains words for a specific language, where each word has its own phonetic representation and frequency grade to make sure the word is accurately recognized.

A Real-Time Speech Analytics Model is a zip file containing language modeling files and a set of fundamental rules for how the Real-Time Speech Analytics Engine should recognize audio.

The Real-Time Speech Analytics Model file also includes metadata xml files. The metadata xml files contain information about the Real-Time Speech Analytics Model, such as the language and vocabulary it supports, and its version number.

Supported Real-Time Speech Analytics Models

You must use a Real-Time Speech Analytics Model that is designed for version 15.2 (and higher) servers, and is designed for use with the Real-Time Speech Analytics feature. You cannot use a Real-Time Speech Analytics Model that was designed for releases earlier than version 15.2 with a 15.2 (or higher) server, or one that was designed for use with the standard speech analytics feature.

An error message appears in the user interface if you attempt to upload a Real-Time Speech Analytics Model designed for an earlier release to a version 15.2 (or higher) server. An alarm indicating the language model is obsolete is raised if you attempt to manually copy a Real-Time Speech Analytics Model designed for an earlier release to a directory on a version 15.2 (or higher) server.

Obtaining Real-Time Speech Analytics Models

Contact your customer support representative to obtain a Real-Time Speech Analytics Model that is supported by the version 15.2 (and higher) servers.

Verint personnel and business partners can obtain a language model by sending an email request to the mailbox named Speech.Enablement.The request must include the customer name and line of business, the language and vocabulary of the language model, and the analytics model type as Real-Time Speech Analytics Model, (for example: Acme corporation, telco, English US, Real-Time Speech Analytics Model).

Attributes of Real-Time Speech Analytics Models

The three most important attributes of a Real-Time Speech Analytics Model are language, vocabulary, and version. These attributes are discussed below:

Language - A Real-Time Speech Analytics Model supports one language. For example, a model supporting English is a different model from one supporting Spanish.
Vocabulary - A model supports one vocabulary of one language.

Some languages are spoken differently in different areas of the world. Different vocabularies are required to accurately model different dialects of a language.

For example, two of the supported vocabularies of the English language are:
- English - United States (US)
- English - Great Britain (GB)
Models that support the same language but different vocabularies of that language are different models. For example, the English - US model is a separate model from the English - UK model. To use both of these models with Real-Time Speech Analytics, you have to upload two separate model files.
Version - Models are periodically updated.

The version indicates the supported language, vocabulary, and version number of the language model.

An example format for a language model version is:

en_US-7

In this example:
- en indicates the supported language (English in this example)
- US indicates the supported vocabulary (United States in this example)
- 7 is the version number of the language model. The version number of a model is incremented when there is a new release of the model.
Other examples include:
- en_GB-9 - Lanuage is English, vocabulary is Great Britain, and version is 9.
- fr_FR-4 - Language is French, vocabulary is France, and version is 4.