What is SSML? Pauses, pronunciation, and supported markup
Understand speech-synthesis markup through a standard XML pause example, distinguishing specifications, vendor extensions, voice support, and listening checks.
Understand the concept
SSML uses XML-based markup to request pauses, pronunciation, and related rendition from supported TTS systems. Providers and voices support different subsets and extensions. It is not audio, and valid markup does not guarantee natural results. Confirm support and listen rather than treating tags as synthesis evidence.
Markup expresses pauses rather than spoken tags
The example requests a 300-millisecond pause between English sentences. It illustrates standard semantics rather than a universal complete API request; services may also require voices or fields. Plain-text inputs may ignore or speak unsupported tags, so check input type.
Escape XML-special characters so user text does not break structure. Valid tags do not establish voice support. W3C semantics and provider subsets or extensions require separate checks.
<speak version="1.1" xmlns="http://www.w3.org/2001/10/synthesis" xml:lang="en-US">
Hello.
<break time="300ms"/>
Welcome.
</speak>Check pronunciation and prosody controls by language
Dates and digit-by-digit readings differ in meaning, and rate or pitch settings can affect naturalness. Microsoft documents structures and extensions with voice and language limits. Do not transfer support assumptions across services.
Listen to names rather than only inspect tag spelling, and recheck after voice updates. No synthesis API was called or commercial voice timing validated here.
Listen and inspect audio properties after synthesis
Test pauses and pronunciation separately on short sentences before combining long scripts. Inspect missing words, unusual pauses, and editing-compatible formats. This is a checking flow without a universal speaking rate.
SSML controls rendition, not factual truth or voice authorization. Correct dates and numbers first. Keep text and markup versions for diagnosis; playable output still needs listening.
Checked sources
These sources support the concepts and methods described here. Examples and test scope are labeled separately; check the version of the tools you use.