Understanding SSML

Introduction to SSML

SSML operates by augmenting the plain text that is passed to a synthesizer with a predefined set of elements, or tags. An XML parser first separates the plain input text from the markup specifications. The specifications are then processed and sent as a set of instructions in a form that can be understood by the synthesizer to produce the desired effects. For the XML parser to carry out this job, the markup needs to be well formed; for example, elements must be closed and multiple elements must be properly nested. For an introduction to basic XML concepts, see w3schools.com/xml/xml_whatis.asp.

An SSML element is anything contained within, and including, an opening tag and its matching closing tag. As shown in the following example, an element can contain a combination of other elements (tags can be nested) and text. Additionally, elements can require or optionally accept attributes set to particular values.

<tag1>
  <tag2 attributeName="attributeValue">
    ... some text ...
  </tag2>
</tag1>

A full legal SSML document consists of an XML prolog, which contains information such as encoding and the schema against which to validate the SSML document, followed by the root element, <speak>. Within the span of the <speak> element, you specify the text that is to be synthesized, augmented with additional elements.

<!-- The XML Prolog -->
<?xml version="1.0" encoding="UTF-8"?>

<!-- Root Element -->
<speak version="1.1">
  ... the body that contains text to be synthesized plus markup ...
</speak>

The XML prolog is not needed for text that you pass to the service. The service supports SSML fragments, which are SSML elements that do not include the full XML header and do not need to include their parent elements. For example, the <?xml> and <speak> elements are always optional for SSML that you send for synthesis.

SSML support

For more information about using SSML and related features with the service, see the following:

• The service implements most of the W3C specification and supports SSML fragments.
  • For complete information about the service's level of support for all SSML elements, see SSML elements.
  • For examples of using SSML elements with the text of a speech synthesis request, see Examples of input text.
• The service supports additional synthesis features for enhanced neural and expressive neural voices:
  • Modifying the speaking rate
  • Modifying the speaking pitch
• The service supports additional synthesis features for expressive neural voices:
  • Using speaking styles
  • Emphasizing interjections
  • Emphasizing words
• The service supports a synthesis feature that lets you control how alphanumeric strings are spelled out for German voices. For more information, see Specifying how strings are spelled out.
• The service supports the use of the SSML <mark> element with the WebSocket interface to obtain timing information for words of the resulting audio. The WebSocket interface also allows you to request information for all strings of the input text. For more information, see
  • Specifying an SSML mark
  • Requesting word timings for all words
• The service's customization interface supports the use of the SSML <phoneme> element to specify the phonetic spelling that it uses to pronounce a word. The phonetic spelling represents the sounds of a word, how those sounds are divided into syllables, and which syllables receive stress.
  • For information about the customization interface, see Understanding customization.
  • For information about the valid symbols that you can use in an International Phonetic Alphabet (IPA) or the IBM Symbolic Phonetic Representation (SPR) specification for any supported language, see Understanding phonetic symbols.

SSML validation

The service validates all SSML elements that you submit in any content, either as input text for synthesis or as the definition of a word's translation for customization. The service cannot determine ahead of time whether text submitted for synthesis contains SSML elements. Therefore, it performs the same validation for all input text, regardless of whether it contains SSML.