SSML or plain text: keep speech controls out of unsupported workflows

Speech markup can express pronunciation or pauses, but only when the chosen synthesis engine understands the requested elements. A script field that accepts text is not automatically an SSML interface, and visible tags can become spoken or rejected content.

Make the decision explicit

Keep the editorial narration separate from provider-specific speech controls. Start from a reviewed plain-text version. Add markup only in a workflow with documented support, and preserve both representations. Check whether the provider counts markup toward limits and whether a voice or model imposes additional restrictions. Do not assume support from another provider's example.

A worked example

A fictional script contains the initials AB. The editorial file records the intended spoken form as two letters. A compatible engine might accept a spelling control, while a plain-text engine may require a different input convention. Compare the actual test recording with the intended result. The important artifact is the approved pronunciation, not the presence of a particular tag.

Review before handoff

After generating audio, prepare captions from the intended spoken words rather than exposing speech-control markup to viewers. Keep the engine settings with the recording, and retest after changes. Topicut's free text estimator does not execute SSML; using it to count a markup-heavy document without inspection can misrepresent the spoken content.

Put the idea to work

Read ssml and use the related free workflow tool for its stated, limited task. Neither link replaces the review described above.

Reference and editorial notes

• Google Cloud: supported SSML controls

References checked October 29, 2026. Numerical and fictional scenarios are illustrative unless explicitly identified otherwise. The cover is an original Topicut explanatory diagram, not a product screenshot or a measurement.