A voice style is a production choice: the combination of perceived energy, pace, accent, vocal age, warmth, and authority that supports a particular script. It is not a shortcut for guessing what an audience “should” sound like.

By the end, you will have a three-niche voice test sheet: three blind auditions, exclusions, scores, and reproducible choices.

// concept

Turn the Niche Into a Voice Brief

Start with the listener’s task. A finance explainer must make figures easy to follow; a horror story needs controlled tension; children’s education needs clear, safe instructions. These requirements do not make one gender, accent, or age inherently more trustworthy.

Use this matrix as a starting hypothesis, then test it with real listeners:

Content niche	Audience and trust need	Energy	Pacing	Accent / vocal-age decision	Exclude before scoring
Personal-finance explainer	Listener must distinguish facts, examples, and cautions	Calm, alert	Medium; pause around figures	Use the audience’s understood language; vocal age only if the brand brief requires it	Rushed numbers, sales-pitch excitement, swallowed disclaimers
Horror story	Listener must follow plot while tension builds	Restrained, rising at turns	Slow-to-medium with deliberate silence	Accent should serve setting or narrator identity, never caricature it	Cheerful cadence, constant whispering, uncontrolled volume jumps
Children’s science lesson	Listener must understand steps and remain engaged	Warm, lively, not loud	Medium with short sentences	Choose for clarity and safeguarding; do not imitate a child to manufacture trust	Baby-talk, shouting, unclear safety instructions

Age, gender, and accent are not quality scores. Specify them only when the character, language, accessibility test, or approved brief requires it.

In ElevenLabs, the documented route is Voices → Explore, with language, accent, category, gender, and age filters. Confirm the current route and plan limits. Shortlist three authorised voices.

// concept

Lock the Voice Without Locking Every Performance

Save the voice ID, model, settings, pronunciation notes, and approved reference clip. Direct emotion and pace without changing identity.

Changing narrators can weaken audience familiarity. Label temporary character voices; retest permanent changes when language, audience, or content type shifts.

// worked_example

Worked Example

This completed sheet is labelled sample data for a hypothetical Islamabad creator. It does not report real audience performance. Each niche used its paragraph unchanged for A, B, and C:

// prompt — copy me13 lines

FINANCE — Sample only: If a phone costs one hundred and twenty thousand rupees, a
twelve-month plan is not automatically cheaper. Compare the full payable amount,
processing fee, and late-payment terms. Then check whether the instalment still fits
after rent, travel, and an emergency reserve.

HORROR — Sample only: At midnight, Hamza heard the upstairs tap running. The house
had no upstairs floor. He switched on his phone torch, and the water stopped. Then
one wet footprint appeared beside his own—facing the other way. From the ceiling,
someone whispered his name.

CHILDREN’S SCIENCE — Sample only: Place one paper towel in each cup. Pour two
spoonfuls of water into the first cup and four into the second. Predict which towel
will absorb more. Keep the cups on a flat table, wipe spills, and ask an adult to help.

Settings: Eleven v3; Stability: Natural; identical format; no tags; default speed. Confirm current controls in the official guide.

Scores below are I/T/E/P: intelligibility, trust fit, energy/pace, pronunciation.

Niche	A	B	C	Locked choice and note
Finance	5/2/2/5; salesy	3/4/4/2; amount blurred	5/5/4/5	C; pause after sentence one
Horror	5/2/1/5; cheerful	4/5/5/4	3/3/5/4; volume jumped	B; keep restrained opening
Children’s science	4/3/2/5; baby-talk	5/5/4/5	3/4/5/3; “absorb” unclear	B; preserve calm safety line

The bad finance draft used A because its demo sounded polished, but rising emphasis made caution sound like a sale. The fix was not “more professional”: C won the blind test, and a paragraph break created the pause. The sheet locked C’s identifier, model, setting, and “rupees” pronunciation.

// failure_cases

Failure Cases to Diagnose

6 cases to diagnose

Demo-reel bias
a prepared showcase wins. Test your own difficult paragraph.
Changing two variables
candidate B has different stability or speed. Reset every setting so only the voice changes.
Accent stereotyping
a reviewer calls an accent “untrustworthy” without citing comprehension. Blind the labels and require timestamped evidence.
Expressiveness hiding errors
dramatic delivery masks a number or name. Review with the transcript open.
No exclusion gate
the highest average wins despite unclear safety language. Apply rights, intelligibility, and pronunciation gates first.
Voice drift
the editor follows every trend. Store the voice lock with the project template.

// pakistan_angle

Pakistan Angle

Test Pakistani English, Urdu, and Roman-Urdu with rupee amounts, lakh/crore, Rawalpindi, Faisalabad, and easypaisa. A global demo may misread them. Ask a Pakistani listener to score comprehension on a phone speaker; never rank a Pakistani accent lower merely because it differs from British or American English.

For Urdu or Roman-Urdu, save approved phonetic spellings and audition code-switched sentences. Never clone staff or public figures from WhatsApp audio. Use your own or an authorised library voice, or obtain documented permission specifying commercial use. Disclose synthetic narration where required.

// hands_on

Hands-On Exercise

5 steps

Build a three-niche voice test sheet for finance explainer, horror story, and children’s lesson:

Complete the matrix fields for audience, trust need, energy, pace, language/accent rationale, and exclusions.
Write one 45–70-word test script per niche, including its hardest pronunciation or delivery moment.
Audition three voices per script with identical model settings. Name the exports only A, B, and C.
Have one listener score the four criteria without seeing the voice labels. Record timestamped reasons for scores below 4.
Select or reject each candidate. For every selected voice, record its identifier, model, settings, pronunciation note, and reference-clip filename. Done means three briefs, nine scored clips, explicit exclusions, three choices, and reproducible settings.

// completion_rubric

Completion Rubric

5 checks — tick as you verify

0/5

// sources

Sources

4 official sources — check every claim yourself

// check_yourself

Check yourself

4 questions · answers and options are taken word-for-word from this course

0/4

1 / 4 · diagnose
Your work shows this failure mode: “Demo-reel bias.” The lesson describes it like this: “A prepared showcase wins.” What does the lesson tell you to do about it?