AI voiceover for video: what it actually costs per finished minute

October 28, 2026 · 8 min read · By Resona Editorial Team · Resona Blog

Nobody buys voiceover by the character. Producers buy it by the minute. A script is "four minutes of narration," a course is "nine hours," a game is "two hundred barks." When you compare an AI voice engine against a studio actor, the only number that survives the conversation is the finished minute — and most pricing pages are written in a language that does not translate.

This is the translation. What a finished minute is made of, how a pay-as-you-go model lands against actors and subscriptions, and where the costs that never appear on the pricing page actually hide.

What a "finished minute" is made of

A finished minute of voiceover is not one minute of raw generation. It is the minute after:

Clean capture. No clicks, no room tone bleeding through, consistent distance from the mic.
Correct pacing. Pauses in the right places, emphasis on the right words — a punctuation and chunking problem more than a model problem.
Consistency across the whole piece. A ten-minute explainer has to sound like one person in one room, not ten generations stitched together.
The revision pass. The line that reads flat on camera, the number that was misread, the client note "warmer, please."

When you price voiceover, you are pricing the minute after all four of those, not the minute off the first render. Any honest cost comparison has to include the revision pass, because that is where the two models diverge most.

The three ways voice is priced, and what each hides

1. Hiring a voice actor

A working narrator runs roughly $200–400 per finished hour, which is about $3–7 per finished minute once you include studio and direction. The price is transparent and the quality ceiling is the highest. The catch is throughput: every revision is another session, and a nine-hour course is a project, not a task. For a one-off hero video, this is often the right call.

2. A monthly subscription TTS plan

Subscriptions quote a per-character or per-minute allowance inside a fixed monthly fee. The hidden line is the allowance itself: the moment you exceed it, you are buying a higher tier, and the tier you need may not exist. Subscriptions also tend to attach rights that expire when you stop paying, which matters if the videos stay online after the plan lapses.

3. Pay-as-you-go, billed on what you generate

A per-byte or per-credit engine charges for what you actually render, with credits bought in advance and never expiring. No monthly commitment, no tier cliff. The thing it does not hide is the per-unit rate — you can compute the exact cost of any script before generating it. The trade is that there is no "flat fee" safety net if you suddenly produce a lot.

Turning per-byte billing into per-minute

On a per-byte engine, speech is billed at 4,200 credits per million UTF-8 bytes; English is one byte per character, and one finished minute of narration is roughly 900 characters. That is about 3.8 credits, or about $0.08, per finished minute of English narration — before any voice clone. A cloned voice adds a one-time 100 credits (about $2) that you reuse for every minute after. A ten-minute video lands around $0.76 plus the clone, and the credits never expire.

The number looks small because it is. The reason it matters is not the per-minute rate — it is what happens on revision three.

Where the real cost hides: revisions

This is the part of the comparison that flips the answer.

With an actor: a "can you redo this line warmer" note is another booking. Three revisions on a ten-minute video can double the cost.
With a subscription: revisions are included until they are not — they eat the monthly allowance, and a big revision pass can quietly push you into the next tier the following month.
With pay-as-you-go: every revision is just more generated audio at the same per-minute rate. The third revision costs the same as the first. This is the structure that favors iteration, and iteration is most of what voiceover actually is.

If your work is "record once, ship," an actor is competitive. If your work is "generate, listen, adjust, regenerate," the per-render cost dominates, and pay-as-you-go wins by construction.

The other hidden line: languages

A single script in five languages is five narration sessions with actors, or five plan upgrades with subscriptions that cap non-English minutes. With a per-byte engine and a cloned voice, it is the same voice identity rendered in each language, billed at the same per-minute rate per language. The cost scales with minutes, not with how many markets you serve — which is the difference between "we localized the trailer" being a budget line and being a checkbox.

Where AI voiceover is clearly the right tool

Explainer and product videos. One register, one speaker, heavy revision. The exact shape where per-render cost helps.
E-learning and courses. Hours of consistent narration where hiring scales badly and a cloned voice keeps the whole course sounding like one instructor.
IVR, prompts, and notifications. Short, high-volume, frequently updated. No actor books a session to re-record a 12-second menu option.
Game NPC barks and localization. Hundreds of short lines, often in several languages, where consistency matters more than a single virtuoso performance.

Where it is not, yet

A character-led film trailer, a brand hero voice expected to stay identical for a decade, or any piece where a single emotional performance carries the whole spot — those are still actor territory, because the value is the performance, not the throughput. Being honest about this is the difference between a tool that saves money and one that quietly produces something you cannot ship.

A sample budget, side by side

Ten minutes of English explainer narration, three revision passes, one cloned voice:

Option	First pass	Plus 3 revisions	Rights after you stop
Voice actor	~$50–70	often 1.5–2×	per contract
Subscription tier	included (until cap)	eats monthly allowance	may lapse
Pay-as-you-go	~$0.76 + $2 clone	same per pass	credits owned

The table understates the actor's real cost only if revisions stay low, and overstates the subscription only if you never exceed the cap. The pay-as-you-go column is the one that does not move when you iterate.

What to check before you commit to any engine

Does the rate survive revision? Flat per-render pricing is worth more than a low headline number with a hidden cap.
Do the rights outlive the plan? If you stop paying, can the videos you already published keep their audio?
Is the per-minute cost computable before you generate? If you cannot price a script from the pricing page, the page is hiding something.
Does it localize without rebooking? If your audience is multi-market, the language math decides the budget.

The honest summary

AI voiceover is not cheaper than an actor for a single, perfect, emotionally led spot — and pretending otherwise produces videos you do not ship. It is dramatically cheaper the moment the work becomes iterative, multilingual, or high-volume, because those are exactly the cases where the actor model's per-session cost and the subscription model's cap both break down. Price it in finished minutes, include the revision pass, and the right tool for a given video becomes obvious instead of ideological.

To price your own video, generate a minute in the Resona console and read the credit cost before you commit. Speech is billed by the byte, packs start at $9, and credits never expire — see the pricing page.

New to this topic? Read the full Text-to-Speech Guide.

Sources

Wikipedia — Voice acting