Why Two Quotes for the Same Script Look Nothing Alike
Pick a script, take it to three AI voice tools, and you will get three prices that share no common unit. One counts characters, another counts minutes of finished audio, and a third simply sells you a seat.
None of them are hiding anything. They just measure the same work differently, which makes side-by-side comparison impossible until you convert everything into one unit.
The mismatch matters most at the two extremes. Short scripts rendered often behave nothing like long narration produced once a month, and the cheapest plan flips between them.
This guide converts the units, shows what silently consumes quota, and sets out which plan shape suits which kind of project.
The Four Units Vendors Actually Bill In
- ● Characters favour short scripts
- ● Minutes favour steady narration
- ● Seats favour teams, not volume
Characters are the oldest unit and still the most common on developer platforms. The engine counts the text you submit, then charges for it whether or not you keep the result.
Minutes of generated audio feel more intuitive for video work. A plan quoted in minutes maps directly onto a finished timeline, which makes budgeting simple for narration.
Credits sit in between and cause most of the confusion. A credit usually maps to a fixed number of characters, but premium voices, cloning, and dubbing often consume credits at a higher rate.
Seats are the fourth model, common in editing suites where voice is one feature among many. Volume limits still exist underneath, so a seat rarely means unlimited generation.
Converting Your Script Into Any of Them
One conversion chain covers almost every estimate you need. English averages close to six characters per word once you include the trailing space.
Speech adds the second link. Narration for video usually lands between one hundred and forty and one hundred sixty words per minute, so a minute of audio sits near nine hundred characters.
Work the chain in whichever direction your quote demands. A two thousand word script becomes roughly twelve thousand characters and roughly fourteen minutes of finished audio.
Then add a correction margin before comparing plans. Most projects render 20% to 50% more audio than the final cut contains, because takes get replaced.
How the Pricing Models Compare
- ● Match the unit to your workload
- ● Check overage before you subscribe
- ● Free tiers rarely allow monetising
The table below describes the shapes rather than the numbers, since rates change frequently. Confirm current pricing on each vendor’s official page before committing, and note the terms as of 2026.
| Billing model | Typical examples | Best for | Where it hurts | Watch out for |
|---|---|---|---|---|
| Per character, pay as you go | Google Cloud Text-to-Speech, Amazon Polly, Azure | Irregular or automated workloads | Long-form narration at premium voice tiers | Markup counted as characters |
| Character or credit subscription | ElevenLabs and similar studio tools | Steady monthly output | Quiet months where credits expire | Premium voices spending credits faster |
| Minutes of generated audio | Murf and similar video-first tools | Video narration with a known runtime | Heavy re-recording during edits | Whether drafts count toward minutes |
| Per seat with usage caps | Descript and other editing suites | Teams who need editing plus voice | Solo creators paying for collaboration | Caps buried under the seat price |
| Free tier | Most vendors offer one | Testing voices and pronunciation | Anything published or monetised | Attribution rules and commercial limits |
| Enterprise or custom | Larger platforms on request | Predictable high volume | Small projects with no leverage | Annual commitments and minimums |
Two rows deserve a second look. Per-character API billing rewards automation and punishes long narration, while minute-based plans do the opposite.
The vendor pages themselves remain the only reliable source for rates. The ElevenLabs pricing page and Google Cloud Text-to-Speech pricing both show how differently two credible platforms structure the same service.
What Quietly Burns Your Quota
- ● Re-renders cost full price
- ● Markup can count as characters
- ● Long blocks make edits expensive
Re-rendering is the biggest hidden cost. Fixing a single mispronounced surname usually means regenerating the whole block that contains it, at full price.
Markup counts as text on several platforms. Pause tags, emphasis, and phonetic spellings all add characters that the listener never hears but the meter still records.
Voice previews add up during casting. Auditioning a paragraph across a dozen voices burns twelve times the quota of the final render.
Premium voice tiers multiply everything. The most natural voices frequently consume credits at a higher rate, so a plan sized on the standard rate runs dry early.
Dubbing and translation sit in their own category. Producing one script in five languages costs five times the base, plus whatever the translation layer charges.
Overage behaviour decides how bad a bad month gets. Some platforms stop generating once the quota is gone, while others keep going and bill the difference, and only one of those surprises you.
Numbers and abbreviations cause a subtler kind of waste. A figure read as digits instead of words forces a rewrite of the surrounding sentence, so writing out awkward terms in the script saves several rounds of regeneration.
Team accounts add one more leak. Shared credit pools drain quickly when several people audition voices in the same week, and most tools show usage only after the fact.
A Worked Example You Can Copy
Take a weekly ten-minute episode as the test case. 10 minutes of narration is roughly fifteen hundred words, or nine thousand characters before markup.
Four episodes a month brings the baseline to thirty-six thousand characters. Add a 50% correction margin for retakes and pickups, and the planning figure becomes fifty-four thousand.
Now translate that into each unit. A character subscription needs a tier above fifty-four thousand, a minutes plan needs roughly 60 minutes, and an API bill scales with the same character count without any monthly floor.
The comparison flips as volume changes. One episode a month usually favours pay-as-you-go, while a daily show favours whichever subscription tier has room to spare.
Run the same arithmetic on your own worst month rather than your average. Plans that fail do so during the busy weeks, and a quota that runs out mid-project costs more than the tier you avoided.
Pricing That Script on Real Meters
Now hold the 54,000-character planning figure against published rates. Prices are as of October 2026, and the example assumes they stay there.
| Option | Published rate | Result for 54,000 characters a month |
|---|---|---|
| ElevenLabs Starter | $6 a month for 30,000 credits, 1 credit per character on Multilingual v2 | 24,000 characters short |
| ElevenLabs Creator | $22 a month for 121,000 credits | Fits, with 67,000 credits spare |
| Amazon Polly neural voices | $16.00 per 1 million characters | 54,000 × $16 ÷ 1,000,000 = $0.86 |
| Amazon Polly generative voices | $30 per 1 million characters | $1.62 |
| Murf API studio voices | $0.03 per 1,000 characters | $1.62 |
The Polly neural figure is rounded to the cent from $0.864. Polly’s free tier also covers 1 million neural characters a month for the first 12 months, so this workload would cost nothing there in year one.
Over a full year the gap looks dramatic. Twelve months of ElevenLabs Creator at the monthly price is $264, against about $10.37 on Polly neural voices once the free period ends.
The subscription is not simply overpriced, though. It buys an editor, a voice library, and a commercial licence, while an API returns raw audio and leaves the production work to you.
So per-character APIs win when software assembles the audio, and subscriptions win when a person edits every episode by hand.
Rights, Cloning, and the Clauses That Cost More
Commercial use is a licensing question rather than a technical one. Several providers grant it only on paid tiers, and free output can carry attribution requirements.
Voice cloning usually sits above the standard tier. Instant clones from a short sample and professional clones from longer recordings are priced differently, and consent rules apply to both.
Ownership language varies more than buyers expect. Some vendors grant broad rights to generated audio, while others reserve rights that matter for advertising or broadcast.
Cancellation is the clause people read last. Access to previously generated files sometimes ends with the subscription, so downloading finished audio promptly is simple insurance.
If you are weighing cloning against a stock voice, our comparison of voice cloning and text to speech covers the quality and consent trade-offs behind these tiers.
Which Plan Shape Fits Your Project
The weekly podcast producer: Minutes-based plans match your workload, since runtime is predictable and edits are modest. Check whether draft renders count against the allowance.
The developer adding narration to an app: Per-character API billing is the natural fit. Usage scales with real users rather than a guess made in advance.
The course creator with one big launch: Buy a large plan for one month rather than a small plan for twelve. Credits rarely roll over, so concentrating the work saves real money.
The marketing team producing many short spots: Watch the re-render rate rather than the headline quota. Short scripts revised repeatedly consume far more than their word count suggests.
The multilingual publisher: Price the dubbing pipeline separately from narration. Translation and per-language rendering usually dominate the total.
The occasional user: Stay on pay-as-you-go and skip subscriptions entirely. A dormant monthly plan is the most common form of waste in this category.
Estimating a Project Before You Subscribe
Start from the finished runtime you need, not the script you have. Multiply by nine hundred characters per minute to get a baseline in the most common billing unit.
Add a correction margin next. A first-time project should assume half again as much rendering, and a polished team can plan on a quarter.
Then price that total under two different models before choosing. The comparison usually separates the plans by a wide margin, and the winner depends on your workload shape rather than the vendor’s reputation.
Finally, run a real page of your own script through the free tier of each shortlisted tool. Pronunciation of names, numbers, and technical terms varies enormously, and our roundup of AI voice generators is a starting point for that shortlist.
Cost per character only matters once you have settled on a voice you can stand to hear repeatedly. How to pick an AI voice you can live with covers the listening tests worth running before you commit a plan to a whole series.
Character counts cover the markup you add too, which matters once scripts carry tags. What SSML tags do in an AI voice tool explains which ones are worth the characters they cost.
FAQ
How many characters is a one-thousand-word script?
Roughly six characters per English word, including the space that follows it. A thousand-word script therefore lands near six thousand characters before any markup. Spoken at a normal narration pace, that same script runs close to seven minutes.
Do SSML tags and pause markup count toward my character quota?
On several platforms yes, because the billing counts every character sent to the engine rather than the words the listener hears. Pause tags, emphasis tags, and phonetic spellings all add characters. Check the documentation for your provider before you mark up a long script.
Does regenerating the same line charge me twice?
Every render counts on nearly every platform, so fixing one mispronounced name in a long paragraph can cost the whole paragraph again. Splitting scripts into short blocks keeps corrections cheap. Some tools now regenerate a single sentence, which is worth looking for.
Can I use the audio from a free plan in a monetised video?
Not always, since several providers restrict commercial use to paid tiers and require attribution on free ones. Voice cloning often sits behind a higher tier again. Read the licence terms on the official site for the specific plan you intend to buy.
Do unused voice credits roll over to the next month?
Usually not, because most subscription credits reset each billing cycle rather than rolling over. A heavy month followed by a quiet one wastes the quota you paid for. Pay-as-you-go API billing suits irregular workloads better.
Sources
- ElevenLabs pricing — checked 2026-09-27
- Amazon Polly: Pricing — checked 2026-10-03
- Murf AI: Pricing and plans — checked 2026-10-03
Some links may be affiliate links. We may earn a commission at no extra cost to you.
This article was written with AI assistance. It is researched and fact-checked, not based on personal hands-on testing unless explicitly stated.
Comments
Post a Comment