Skip to content
Get in touch

Give words a voice

Explore text-to-speech (TTS), human recording, editing, and permitted voice customization for narration and spoken audio.

Go to My experiment

Explore

What are the ways to do this?

Two illustrative readings of one short excerpt

Read the same roughly 25-second excerpt in a clear style and a deeper, more measured style. Keep the words fixed; listen for pace, pronunciation, and fit.

Notice: What changes when the words stay the same but the delivery changes?

Clear voice — same excerptA synthetic voice reading the 21-second invitation excerpt. This is an illustrative style comparison, not the full experiment passage.

Credit: Created for Renaissance Solutions with AI Doc Maker’s AI Voice Generator (Level 2 Labs), using a stock synthetic voice. Credit details

Deep voice — same excerptA second synthetic voice reads the identical excerpt. Listen for pacing, emphasis, and fit to your audience.

Credit: Created for Renaissance Solutions with AI Doc Maker’s AI Voice Generator (Level 2 Labs), using a stock synthetic voice. Credit details

Read or download the transcript and description.

What this shows: A small, listenable contrast between two generated stock-voice readings made with AI Voice Generator.

What it cannot show: Long-form consistency, all pronunciation controls, permission for customized voices, or performance in another project.

  • These excerpts demonstrate delivery differences, not the full 90-second experiment.
  • They do not establish which product or voice is best.
  • The full passage below includes an acronym, dialogue-like instructions, and an unusual name.

Approach map

Open any approach to inspect its starting material, human work, outputs, and sequence.

Synthetic narrationTurn text into speech with a stock synthetic voice, often called text-to-speech or TTS.
Starts with
A finished or nearly finished script · Pronunciation and pacing notes
Main actions
Choose a permitted stock voice · Generate a reading · Correct pronunciation and pacing · Export and listen in context
Outputs
An audio file · A repeatable script and settings record
Human work
Write for listening · Direct the reading · Check every important word · Edit or regenerate sections
Useful when
You need quick narration iterations · Recording a person is impractical · A disclosed synthetic voice fits the project
  1. Prepare the scriptWrite for ears, mark names, numbers, pauses, and emphasis.
  2. Choose a permitted voiceUse a stock or explicitly authorized voice and record the choice.
  3. Correct the readingChoose the smallest correction that fits the problem.
    • Change the text or pronunciation cueRegenerate a small section: Keep the rest stable and compare the changed passage.
    • Edit the audioTrim or assemble: Use an editor when a clean cut is more reliable than another generation.
  4. Export and listen in placeCheck the actual player, device, captions or transcript, and final format.
Record a human voiceCapture a speaker whose delivery and permission fit the project.
Starts with
A script or speaking outline · A willing speaker · A quiet-enough place and recorder
Main actions
Agree on use and permission · Rehearse difficult passages · Record several takes · Select a complete usable take
Outputs
Original recordings · A selected reading
Human work
Perform or direct · Manage the room and microphone · Listen for meaning and mistakes
Useful when
A personal or accountable voice matters · Natural improvisation helps · The speaker owns the words
  1. Agree on the useRecord who is speaking, where the audio may appear, and how long permission lasts.
  2. Rehearse what mattersMark names, numbers, safety details, and places to breathe.
  3. Record more than one takeKeep the setup stable and leave room tone for editing.
  4. Select and checkListen in the intended setting before deciding the recording is finished.
Record, then editShape a human recording with cuts, cleanup, level adjustment, or assembly.
Starts with
Recorded takes and room tone · A clear edit goal
Main actions
Choose the strongest takes · Remove mistakes without changing meaning · Balance levels and noise · Export a listening copy
Outputs
An edited master · A record of source takes and changes
Human work
Make editorial choices · Protect natural breaths and timing · Avoid over-cleaning
Useful when
The performance is right but needs polish · Sections from different takes must be assembled
  1. Preserve and label takesKeep originals before cutting.
  2. Build a complete readingChoose edits for clarity and meaning.
  3. Make restrained correctionsReduce distractions without making the voice brittle.
  4. Quality checkListen from start to finish on an ordinary device.
Customize a permitted voiceUse voice conversion or a custom model only with explicit, documented permission.
Starts with
A clearly authorized speaker · Approved recordings and terms · A defined use
Main actions
Document consent and limits · Protect source recordings · Create or select the custom voice · Review identity, disclosure, and misuse risks
Outputs
A controlled custom-voice reading · Permission and provenance records
Human work
Obtain and maintain consent · Judge whether the use is respectful · Review every output
Useful when
The speaker has knowingly authorized this exact use · Continuity or accessibility justifies customization
  1. Start with permissionNo technical step replaces informed permission.
  2. Set boundariesDefine allowed material, audiences, storage, deletion, and revocation.
  3. Create and reviewTest on representative material and disclose the synthetic treatment.

Start with your intention

Use My experiment to note your audience, material, limits, equipment, and sharing restrictions. Every field is optional. “Try with our example” fills only an empty goal and constraints field.

Prompt · version 1

Investigate ways to give words a voice

When you know what you want listeners to understand but not how to produce the recording.

Map plausible narration workflows before choosing a product.

Example: A community repair café needs a clear spoken guide for first-time visitors.

Show the full prompt

The published prompt stays unchanged.

Source checked ·

Tools to investigate

A practical starting shortlist, chosen for different workflows rather than ranked quality. Features were checked against official sources on this date; we did not run a comparative test of these services. Plans and interfaces can change.

This shortlist reflects the date above. Confirm current features, terms, pricing, privacy, exports, and accessibility before choosing.

A high-level workflow

  1. Write a short script for one audience and purpose.
  2. Choose a stock synthetic voice or record your own performance.
  3. Compare the same passage, checking names, pauses and emphasis.
  4. Revise, export, and keep the transcript and required credits with the audio.

ElevenLabs

Purpose and fit
A first text-to-speech recording using a stock voice.
How to explore it
Prepare a short script, choose a stock voice, generate a sample, then revise pronunciation and pacing before exporting.
Watch for
Check the plan that applies when you generate: free-tier publication and attribution rules differ from paid commercial use. Keep the script and a copy of the applicable terms.

Sources checked

Descript

Purpose and fit
Narration that you expect to revise alongside an audio or video edit.
How to explore it
Start with a stock AI speaker, generate from your script, edit the words and timing, then review and export the finished piece.
Watch for
A revised sentence can change delivery. Listen to every revision; use only voices you have permission to use.

Sources checked

Audacity

Purpose and fit
Recording your own voice and keeping direct control of the performance.
How to explore it
Record a short take, trim and adjust levels, listen through headphones and speakers, then export an audio copy.
Watch for
This is an audio editor, not a hosted AI voice generator. Keep the editable project as well as the exported file.

Sources checked

Decide

What matters for your project?

Use a few criteria and a separate must-have list. An unknown must-have remains unresolved.

Must-have

Intelligibility

Listeners can follow the meaning without the script.

Acceptable: Every instruction and safety detail is understood on an ordinary device.

Must-have

Pronunciation

Names, acronyms, numbers, and unusual words are spoken as intended.

Acceptable: Niamh, USB, and 10:30 match the supplied reference.

Must-have

Voice permission

Every recorded or customized voice is used with informed permission.

Acceptable: Permission, scope, storage, and disclosure are documented.

Preference

Pacing and expressive fit

The reading gives important ideas enough room and fits the audience.

Acceptable: The welcome feels calm; the safety list remains distinct.

Preference

Correction effort

The number and kind of fixes are visible.

Acceptable: A maker can describe attempts, edits, elapsed time, and remaining problems.

Must-have

Useful export

The file works in the intended player and remains usable later.

Acceptable: A common audio format plus transcript can be retained without a vendor account.

AI reported

What an AI response says. Follow the sources before relying on it.

Source checked

What an opened source states. This does not prove product performance.

We tested

What RS observed in a dated, bounded trial.

Still uncertain

What is unknown, unverified, or outside the trial.

Prompt · version 1.0

Shape voice criteria one question at a time

Before comparing candidates or when “natural” is too vague to judge.

Turn your idea for spoken audio into criteria you can revise.

Example: The guide must pronounce an acronym and an unusual name clearly without sounding rushed.

Show the full prompt

The published prompt stays unchanged.

Prompt · version 1.0

Compare two narration paths without choosing for me

After identifying two feasible narration paths or one path with two settings.

Create an evidence-aware comparison sheet for a small voice trial.

Example: Compare a human recording with light editing against a synthetic reading of the same excerpt.

Show the full prompt

The published prompt stays unchanged.

Experiment

Compare two ways to narrate the same passage

Use one roughly 90-second repair-café passage with two feasible paths, or revise one path once.

Ready-to-use brief

Create a clear spoken invitation for a first-time repair-café visitor. Keep the meaning intact, sound welcoming without sounding promotional, and make the time, repair boundaries, dialogue, acronym, and name easy to hear.

Input

Have you got a lamp that flickers, a jacket with a loose button, or a small object you cannot bear to throw away? Start with a question: could this be repaired? In this imagined community repair café, people bring one object and work beside a volunteer. No experience is needed to begin a conversation. Bring the parts you still have, describe what changed, and leave electrical repairs to someone with the right skills. The café opens at 10:30. The first demonstration lasts 15 minutes. There are three things to remember: bring your object, bring your questions, and allow time to learn. “Can you fix it?” asks a visitor. “Let’s find out what it needs,” says the volunteer. Niamh, whose name is pronounced Neev, has brought a radio. She asks whether its USB cable is the problem. The answer is not obvious, so they check one thing at a time. Some repairs will work. Some will need a spare part or specialist help. Either way, the person leaves knowing more than when they arrived. What would you bring? Pronunciation reference: Niamh = Neev. USB = U S B. 10:30 = ten thirty.

Limit

Aim for 45–60 minutes and no purchase. Set a spending limit before beginning if a paid option is necessary.

Path 1

Record a human reading, then trim pauses and balance the sound.

Path 2

Create a synthetic narration with a permitted stock voice and correct pronunciation or pacing once.

Only one candidate? Make a first reading, change one pronunciation or pacing instruction, then compare the revised reading with the first.

Define success before you begin

  1. Define success before listening to candidates.
  2. Preserve the passage and pronunciation reference unchanged.
  3. Record tool, voice, settings, attempts, elapsed time, and cost.
  4. Listen once for meaning and once for pronunciation, pacing, and expression.
  5. Keep source claims, this trial, and personal observations in separate notes.
  6. Choose a next step without treating one sample as a product ranking.

Observation questions

  • Could a listener understand every number, acronym, name, and safety instruction?
  • Where did the pace help or interfere?
  • How many corrections were needed and how long did they take?
  • Can the final file be exported in a useful format?
  • Does every recorded or customized voice have permission?

What would you do next? Continue, change an approach, check a claim, try a simpler method, or stop.

Recorded on 2026-09-17

We ran the landscape prompt

Compares human recording, text-to-speech, and mixed editing, then proposes a blinded three-arm 90-second passage experiment.

Read the prompt, response, sources, evidence labels, limits, and next question.