ColunTina | Voiceover preparation 1 AUDIO PREPARATION / PRACTICAL GUIDE A voiceover script needs pronunciation notes Separate what should be said from the decisions about how a voice should say it. ColunTina | AI-assisted promotional guide | 17 November 2026 A script can be grammatically correct and still leave a narrator guessing. Acronyms, product names, dates and units may each have more than one plausible spoken form. When those decisions live only in a chat message, a later revision can undo them. A small pronunciation sheet gives the author and reviewer a shared reference. 1. Mark the decisions that are still open Imagine a fictional 45-second introduction to a software demo. The draft mentions SQL, a version number and a made-up project called LumaVale. This is an authoring exercise, not an audio production that has been performed. The name is invented; no real brand pronunciation is asserted. Highlight tokens that need a decision. Does the author want SQL spoken as individual letters or as a word? Is version 2.1 read as two point one, or does the audience use another convention? A model may choose a plausible answer without knowing the intended one. Mark an unresolved choice instead of treating the first generated reading as authoritative. 2. Keep the readable script intact Maintain a clean copy of the intended spoken message, then keep pronunciation instructions in a separate sheet or the supported instruction field. Do not assume that a bracketed note will be interpreted as direction rather than read aloud. The exact instruction syntax depends on the selected voice tool and should be checked in its current documentation. ColunTina | Voiceover preparation 2 WORKED EXAMPLE / REVIEW CARD Record a choice and the evidence for it The following entries are proposed editorial choices for the fictional demo. They are not tested pronunciations, provider-specific markup or recommendations for every audience. Token Proposed direction Status / next check SQL Use the individual letters S, Q and L. Author preference in this example; listen to a sample before approval. 2.1 Read as "two point one." Check that the context clearly identifies a version. LumaVale Ask the project owner for an intended reading. Unresolved; do not invent an official pronunciation. 2026 Choose the form appropriate to the audience and script. Editorial choice pending; keep it consistent after approval. 3. Review a short sample before a long render When the selected tool supports the required output and the task permits it, generate a short sample containing the difficult tokens. Listen for the intended words, pacing, pauses and awkward transitions. Record what was heard and which file was reviewed. A written instruction sheet is not proof that the audio follows it. If a sample has not been generated or heard, say so. The review status can remain "prepared, not auditioned." Do not describe a chosen voice as natural, accurate or approved on the basis of a provider label alone. Keep unapproved samples out of a finished delivery. 4. Measure duration rather than promising it A 45-second target is a constraint to verify in the exported audio. Word count alone cannot establish the result: pronunciation, pauses and speed affect timing. If the sample is too long, revise the wording or pacing deliberately, then check again. Avoid cutting essential meaning simply to match an estimated duration. ColunTina | Voiceover preparation 3 A PROPOSED ORKAS WORKFLOW Keep script, notes and audio review together Orkas is a desktop app for coordinating AI work. A possible process is to give Commander the approved script, audience and pronunciation sheet, ask ContentWriter to refine the wording, and use VoiceStudio for a narration workflow with the tools and voices actually available. This guide does not claim that an Orkas audio task has run. A bounded task could read: "Prepare the fictional demo narration from the approved script. Keep unresolved pronunciation choices visible. Use only a voice we are permitted to use. Produce a short review sample when supported, identify the generated file, and report the duration and checks actually performed. Do not mark the final narration approved without a real listening review." 5. Preserve the revision trail Keep the script version, pronunciation-sheet version and audio filename together. If a term changes, identify which recording uses the new wording. Keep the relevant materials and instructions in a Project, and select useful Library sources when reusing reference documents. Context reuse still requires choosing the right sources and checking the output. 6. Hand off an honest status A useful delivery includes the approved script, pronunciation choices, the playable audio when actually produced, and a short note listing completed and outstanding checks. Separate "script reviewed," "sample heard" and "final approved." These describe different observations. The next person should not have to guess which stage is complete. Use original scripts and voices for which the task has appropriate permission. Do not imitate a real person or use a recording without the necessary rights and consent. This example requires no voice cloning, private customer recording or real speaker identity. Orkas has a free core desktop app. Selected model providers, audio tools and optional services may require access or add costs. Working files and context remain on the device by default; remote model and tool calls send relevant content to chosen services, and enabled sharing or sync uploads its selected scope. Exact voice options, formats and editability depend on the configured workflow.