Published 2026-03-20 · Updated 2026-10-22 · MachenVoice
Text-to-speech sounds robotic when the script was written for the eye, not the ear. This guide shows concrete rewrites you can paste into MachenVoice blocks (up to 2,000 characters each) so neural voices breathe, stress the right words, and survive a phone-speaker preview.
Start with a spoken draft, not a blog post
Blog posts rely on headings, parentheses, and semicolons that neural voices treat oddly. Read your paragraph aloud once before you paste it. If you stumble, the TTS engine will too.
For MachenVoice, split at natural stops: one block for the hook, another for the lesson, another for the CTA. Regenerating a single block after a wording tweak is faster than re-exporting a ten-minute wall of text.
Before and after: corporate update
Before: “Q3 EBITDA increased 12% (YoY) despite headwinds; however, ops—esp. logistics—remain challenged.”
After: “In the third quarter, earnings before interest and taxes rose twelve percent year over year. Operations teams, especially logistics, still face pressure.” The after version spells out the acronym once, replaces semicolons with periods, and gives the voice room to pause.
Before and after: YouTube hook
Before: “STOP scrolling!! This AI trick will BLOW your mind…”
After: “Stop scrolling. This workflow saves about ten minutes every time you voice a short video.” Capital letters and stacked punctuation often make neural voices sound stiff or shouty; calm wording plus a period usually tests better in preview.
Numbers, dates, and URLs
Write small numbers as words in narration (“three steps”) unless you verified the voice reads digits cleanly. For years, digits are usually fine: “in 2026” not “in twenty twenty-six” unless you prefer formal tone.
Avoid spelling URLs character by character unless you tested it. Say “link in the description” for YouTube, or “visit our support page” for IVR scripts. If you must read a domain, use short words: “machen hyphen voice dot com.”
Punctuation as performance direction
Commas create short pauses; periods create full stops. Em dashes sometimes produce uneven timing—try splitting into two sentences instead.
Ellipses can sound like the voice is trailing off; use them only when you want that effect. For training modules, explicit cues like “Pause here” work better than punctuation alone.
Preview like a listener, not like an author
Generate a short block, listen on laptop speakers, then on a phone. If consonants blur, shorten sentences. If the voice races lists, break bullets into separate sentences.
Keep the same voice ID while you iterate on wording so comparisons stay fair. When the script is stable, export the project MP3 and move on to video or LMS upload.
E-learning objectives: before and after
Before: “Objectives: (1) define GDPR (2) list DSR types (3) sign attestation.”
After: “In this lesson you will define the General Data Protection Regulation, name three data-subject request types, and complete the attestation at the end.” Dense bullet objectives often sound like a robot reading a slide; one flowing sentence per objective usually tests better in MachenVoice preview.
IVR and on-hold scripts
Phone menus need short, repeatable phrases. Write “Press one for billing” instead of “Press 1 for billing” only after you confirm the voice reads the digit naturally in preview.
For estimated wait times, avoid updating every minute in TTS. Use a stable range (“about five minutes”) or rotate a small set of pre-approved clips. Store holiday overrides as separate MP3 files with dates in your runbook.
Acronyms, brands, and tricky words
Spell out an acronym on first use in each block if learners hear the block alone. For internal codenames, replace them with plain descriptions unless marketing requires the trademark.
If a word is mispronounced, try hyphenation (“micro-soft”), a phonetic rewrite, or a different synonym. Neural voices improve with punctuation more often than with creative spelling.
MachenVoice block workflow
Paste block one, generate, listen, edit punctuation, regenerate. Duplicate the voice settings for block two so tone matches. Use the playlist to hear continuity before export.
When a script grows past 2,000 characters, split at chapter headings or scene changes—not mid-sentence. Filename your exports (`lesson2-intro_v3.mp3`) so video and LMS teams know which audio matches the published text.
Lists, quotes, and dialogue
Numbered lists in one block often sound rushed. Convert each item to a full sentence: “First, open settings. Second, choose notifications.”
For two speakers, use separate blocks or label lines (“Host:”, “Guest:”) and pick distinct voices only if your workflow supports alternating exports. Otherwise keep one narrator voice and simplify dialogue to reported speech.
Tone: formal, conversational, and instructional
Corporate compliance scripts benefit from formal but plain English: avoid idioms that do not translate if you later localize slides. Creator scripts can use contractions and questions because listeners expect a host personality.
Instructional tone works when each paragraph ends with a cue: “Next, you will try this in the lab.” That rhythm helps learners know where one idea ends even without seeing the slide.
Testing checklist before you hand off audio
Read the full playlist once without editing. Note any word the voice mispronounces, any sentence that feels too long, and any block that sounds like a different energy level than its neighbors.
Send stakeholders a two-minute sample plus the full MP3 link. Ask them to comment on pacing only, not music or visuals, so feedback stays actionable for script edits in MachenVoice.
Common platform-specific gotchas
YouTube chapters align better when your script includes spoken transitions: “In the next section we cover exports.” TikTok hooks should land in the first block alone so you can trim video without re-cutting narration mid-sentence.
IVR scripts should avoid jokes that age poorly; neutral professional wording survives quarterly reviews. E-learning should match the slide verb tense: if the slide says “Click Save,” the narrator should say “Click Save,” not “You will click Save,” unless style guides require future tense.
| Problem | Fix |
|---|---|
| Semicolons and nested parentheses | Use two short sentences |
| ALL CAPS emphasis | Choose stronger verbs instead |
| Bullet symbols in one block | One spoken sentence per item |
| Unexpanded acronyms | Spell out on first use |
| Wall of text | Split into 2,000-character blocks |
| Legal slide bullets | One spoken sentence per objective |
| Changing wait-time IVR | Stable wording or rotated clips |
| Mispronounced brand | Phonetic rewrite or synonym |