Research-backed calculator guide

AI Dubbing Cost per Approved Minute: Price Review, Revisions and Rejected Output

Calculate the real cost of an approved dubbed minute by combining platform meters, language count, review, revisions, rejected output, and yield.

A quoted AI dubbing rate is rarely the same as the cost of a minute you can actually publish. A platform may meter source media minutes, target-language minutes, characters, credits, or a monthly subscription. Your team may also spend time preparing scripts, checking terminology, reviewing timing, correcting pronunciation, and regenerating rejected segments.

The useful unit is therefore cost per approved minute, not simply cost per generated minute. It answers the operational question: after all in-scope work, how much did each final language-minute that passed approval cost?

What “approved minute” means

Define the deliverable before comparing tools. An approved minute could mean a target-language audio track with a checked transcript. It could instead require music and effects preservation, lip synchronization, subtitles, on-screen text, accessibility checks, or a particular export format. These are different deliverables and should not share one benchmark without a scope note.

Also define whether the unit is a source minute or a language-minute. ElevenLabs’ official dubbing documentation says its dubbing charge is based on minutes of source media for each language selected, with the exact rate depending on the dubbing model. That means a 12-minute source sent to three target languages creates three separately metered language jobs, subject to that product’s terms. ^[1]

An approved minute is not a third billing unit. It is an outcome denominator. You derive it after measuring the costs and the amount of final content that passed the agreed gate. That distinction prevents a common error: counting every generated or regenerated attempt as output even though several attempts may produce only one approved minute.

The approved-minute formula

Use this worksheet:

approved-minute cost =
(platform and usage charges
 + preparation and translation
 + review and QA labor
 + revision and regeneration charges
 + rejected-output cost
 + coordination and allocated overhead
 + in-scope update cost)
÷ final approved language-minutes

Do not add rejected output as a second fee if its production cost is already in the numerator. Rejected output affects the result because its cost remains in the numerator while its minutes stay out of the denominator. The same rule applies to revisions: count their additional work, but do not count the same finished minute twice.

Track approval yield beside the unit cost:

Approval yield = final approved language-minutes ÷ generated language-minutes

Yield is not a quality score, and it is not a promise about another project. It is a diagnostic for your own sample. A low yield may come from transcription errors, translation problems, voice pronunciation, timing, stakeholder changes, or a stricter review gate. Record the reason instead of treating every rejection as a vendor failure.

Worked cost example

Assume a 12-minute English training video is being released in Spanish, French, and German. The figures below are deliberately illustrative estimates, not vendor quotes. They show how to compare a platform quote with internal work on one consistent basis.

Assumptions:

• The illustrative generation rate is $0.80 per generated language-minute.
• One source-preparation and terminology pass costs $12.
• A reviewer costs $30 per hour and spends 90 minutes on the first review plus repeat checks.
• Two revision rounds regenerate eight language-minutes in total.
• Coordination and file handling are allocated at $18.
• The initial run generates 36 language-minutes, but six are rejected or remain unfinished. The final approved output is 30 language-minutes.

The generation-only figure looks much cheaper: $28.80 divided by 30 approved minutes equals $0.96 per approved minute. That number is useful for isolating platform generation, but it is not the cost of the finished deliverable. Once preparation, review, revisions, and coordination are included, the illustrative cost is $3.67 per approved language-minute.

If the same project approved only 24 language-minutes with no change to the $110.20 numerator, the result would rise to $4.59 per approved minute. If it approved all 36 first-pass language-minutes and required no revision work, the cost would be $2.93 per approved minute under the same labor assumptions. The denominator is not a cosmetic detail; it is the main reason headline and finished-output costs diverge.

How provider billing units change the estimate

Do not convert every product into “per minute” until you know what the minute represents. ElevenLabs documents source-media-minute billing per selected language and says the model determines the exact rate. Its documentation also notes that Dubbing v2 can be edited in the UI, while API transcript editing and audio regeneration are limited to Enterprise plans. That access rule can change your labor model if corrections must be handled outside the API. ^[1]

Other stacks expose lower-level components instead of one dubbing price. AssemblyAI’s pricing page lists pre-recorded transcription by audio hour and its Text-to-Speech API at $15 per 1 million input characters as checked on 2026-11-02. A transcript, translation layer, and TTS call should therefore be modeled separately if you assemble the workflow yourself. The $15 figure is a TTS meter, not a quote for a complete approved dub. ^[2]

Deepgram’s official pricing page likewise bills TTS per 1,000 input characters, with the displayed pay-as-you-go rates on 2026-11-02 ranging from $0.015 to $0.045 per 1,000 characters depending on model. Its STT section bills processed audio by the second or minute depending on product presentation and says normal audio requests do not round up to a block. These are component meters, not a finished-output guarantee. ^[3]

Character-based pricing needs a text assumption. If a 12-minute script produces 18,000 characters in each target language, three languages create 54,000 TTS input characters. At an illustrative $0.030 per 1,000 characters, TTS usage would be $1.62 before translation, review, retries, and any platform minimum. State the character count and rate instead of presenting that conversion as a universal per-minute price.

Subscription plans need the same discipline. Divide the portion of the monthly commitment actually allocated to dubbing by the approved minutes produced during the period. If a plan costs $99 but the team uses only 100 approved minutes, the committed-plan allocation is $0.99 before labor. If it uses 900 approved minutes, the same allocation is $0.11 before labor. Those are utilization scenarios, not vendor promises, and shared credits may serve other products.

A practical quote-comparison worksheet

Ask each provider or internal team to return the following fields for the same pilot:

1. Source duration and language list. Record source minutes, target languages, speaker count, and whether each language is billed separately.
2. Meter definition. Label the charge as source minutes, generated language-minutes, characters, credits, subscription allocation, or a custom quote.
3. Included processing. Identify whether transcription, translation, voice generation, mixing, captions, lip sync, and export are included.
4. Review boundary. State who checks meaning, terminology, pronunciation, timing, tone, accessibility, and brand requirements.
5. Retry rules. Capture whether failed generation is refunded, whether a regenerated segment consumes another unit, and whether edits require a higher plan.
6. Output counts. Report generated, rejected, revised, and finally approved language-minutes separately.
7. Labor and overhead. Record reviewer minutes, project-management time, storage, integration, and any internal tooling allocation.
8. Update policy. Test one source-script change so maintenance does not hide inside the launch average.

For model or component comparisons, the site’s Model pricing comparison can help organize unit definitions. If translation or post-processing also invokes language models, use the AI token cost calculator for token-based portions rather than forcing them into an audio-minute estimate. A broader workflow with several model calls can be cross-checked in the LLM cost calculator.

What to exclude or report separately

Separate costs that do not apply consistently to every option. A lip-sync surcharge should not be blended into an audio-only result. A one-time integration build should be amortized over a declared volume or shown separately from recurring production cost. Internal reviewer salaries can be allocated as fully loaded hourly cost, but the assumption must be visible.

Report taxes, currency conversion, storage, music licensing, voice talent consent, and final mastering separately when they are outside the quote. If the source video changes after approval, classify the work as maintenance or a new version before adding it to the launch cohort. Otherwise, an unusually revision-heavy update can make the initial production comparison look worse than it was, or hide recurring maintenance entirely.

A calculator should also preserve the distinction between a vendor quote and an estimate. Use a quoted rate only when the provider has stated the exact meter and scope. Use an estimate for review time, approval yield, rejected output, character counts, or utilization. Replace estimates with pilot observations as soon as the project produces its own evidence.

FAQ

Is AI dubbing cost per approved minute the same as cost per generated minute?

No. Generated-minute cost measures production activity. Approved-minute cost includes all in-scope work and divides by the final language-minutes that passed the approval gate. Rejected and regenerated attempts increase cost without automatically increasing approved output.

Should a multilingual video count source minutes once or once per language?

Count according to the provider’s meter, then report final output in language-minutes. A 12-minute source dubbed into three languages creates 36 target-language minutes of potential output. ElevenLabs explicitly describes charging per source-media minute for each selected language, but other products can use credits or different allowances, so verify the product terms. ^[1]

Do review and QA belong in the AI dubbing cost calculation?

Yes, when review is required for release. Include reviewer time and repeat checks in the numerator. Keep the approval standard constant across providers; otherwise a tool that receives less scrutiny will appear cheaper without producing equivalent deliverables.

How should rejected dubbed minutes be counted?

Include the cost of generating and reviewing rejected output in the numerator, but exclude rejected minutes from the approved denominator. Track rejection reasons and whether the rejection came from quality, a changed brief, a technical failure, or an unfinished job.

How do credits and subscriptions change the result?

Convert the portion of the commitment used by the dubbing project into a cost allocation, then divide by approved minutes. Show both low-utilization and high-utilization cases. Shared credits, plan limits, watermarks, language multipliers, and overage terms can make the effective rate different from the advertised maximum allowance.

Can component API prices be treated as a complete dubbing quote?

No. STT, translation, TTS, and storage prices cover components. You still need assumptions for text volume, language count, retries, QA, mixing, and approval yield. AssemblyAI and Deepgram publish component meters in hours, characters, and audio time; neither component rate alone proves a finished approved-minute cost. ^{[2] [3]}

Limitations

This method does not predict a universal market rate or guarantee a particular approval yield. The worked example uses illustrative assumptions so the arithmetic is inspectable. Vendor prices, plan allowances, model names, taxes, and feature access can change; the official pages cited here were checked on 2026-11-02, and a quote should be rechecked before purchase. A small clean pilot may also understate the cost of noisy audio, many speakers, visible text, sensitive terminology, or frequent source updates.