Research-backed calculator guide

AI Dubbing Cost by Language Count: Budget Source Minutes × Target Markets

Estimate AI dubbing cost per language by rounding source minutes, multiplying target markets, and separating generation, review, licensing, and shared setup.

If you are estimating AI dubbing cost per language, do not multiply the source video length by one flat price and stop there. First convert the source into billable minutes, then multiply those minutes by the number of target languages. After that, add costs that are shared across the project, such as transcript cleanup, and costs that repeat for each language, such as review or re-recording.

That distinction matters because a 12-minute video dubbed into four languages is not a 12-minute job. If the vendor bills by finished minute per language, it is 48 language-minutes before any rounding or plan allowance. The final invoice may still differ because one provider bills language-minutes, another bills credits, and a third exposes only lower-level speech-synthesis units such as characters or tokens.

What “per language” should mean in a budget

There are three quantities that are easy to confuse:

1. Source minutes are the runtime of the original video. They describe the asset you already have.
2. Target languages are the additional language versions you want. The source language is usually not an additional dub unless you are regenerating it.
3. Language-minutes are the billable production units when a provider charges a finished minute for each target language.

If the source is 4 minutes 10 seconds, rounding rules can change the result. With whole-minute billing, use 5 minutes per language. With second-level billing, use 250 seconds per language. If the provider says it charges actual generated audio rather than rounded source runtime, use that provider’s meter instead of assuming whole-minute rounding.

The practical worksheet is therefore:

rounded_source_minutes = ceiling(source_seconds / 60)
language_minutes = rounded_source_minutes × target_language_count

Do not automatically include the original language in target_language_count. If the brief is “English source plus Spanish, French, German, and Japanese,” the count is four. If the project also asks for a new English voice track, add that as a fifth output only when the vendor charges it separately.

Vendor quote versus underlying model cost

An end-to-end dubbing price may bundle transcription, translation, voice generation, timing, mixing, storage, and a download. A text-to-speech API price is narrower. It may cover only synthesis after you have already supplied translated text and handled audio assembly yourself.

For example, Amazon Polly states that it bills the number of characters converted to speech. Its displayed US pricing, checked November 2, 2026, is $4 per million characters for Standard voices, $16 per million for Neural voices, $100 per million for Long-Form voices, and $30 per million for Generative voices outside the relevant free tiers. ^[3] That is a synthesis meter, not an all-in dubbing quote. You still need transcription, translation, timing, mixing, and quality control if your workflow does not provide them elsewhere.

Google Cloud Text-to-Speech likewise says it counts characters in the input, including spaces and newline characters, and lists separate prices by model. The checked page lists Standard and WaveNet at $4 per million characters after the displayed free allowance, Neural2 at $16 per million, and Chirp 3: HD at $30 per million. It also lists Gemini TTS using text and audio-token meters, with audio tokens corresponding to 25 tokens per second of audio. ^[4] Those units can help estimate a custom pipeline, but they should not be presented as a per-language video-dubbing price without translating the script and assembly work into dollars.

Azure Speech provides another useful boundary. Its pricing page describes text-to-speech as per-character billing, lists Neural speech at $15 per million characters for the standard pay-as-you-go row, and separately lists Video Translation input and output video rates. Microsoft labels the displayed prices as estimates rather than actual quotes and notes that region, agreement, date, and currency can affect the final amount. ^[5] When comparing providers, write down whether the number is for text-to-speech only, video translation, or a managed dubbing service.

For a broader monthly AI budget, the site’s Model pricing comparison can help compare model meters, but it does not replace a vendor’s dubbing quote. A character estimate is useful when you are building the pipeline yourself. A language-minute quote is more useful when the platform handles the whole job.

Worked cost example: a 12:30 video in four languages

Assume a 12 minute 30 second English training video. The requested outputs are Spanish, French, German, and Japanese. Use these assumptions:

• Whole-minute billing, so 12:30 rounds up to 13 minutes.
• Four target languages, with no separately regenerated English track.
• Illustrative AI generation rate: $2.00 per language-minute.
• Shared transcript cleanup and project setup: $20.
• Targeted human review: 15 minutes per language at an assumed $25 per reviewer hour.
• Export and temporary storage allowance: $8.
• Taxes, rush fees, lip-sync premium, music replacement, and major script rewrites are excluded.

The resulting unit views are useful for decisions: $39.25 per target language, $3.02 per language-minute, or $12.56 per original source minute. The last number is not a vendor rate. It is simply the total project estimate divided by the 12.5 minutes of source footage.

If the same job used a plan with 40 included language-minutes, only 12 language-minutes would be outside the allowance. If the overage were $2.00 per minute per language, the overage line would be $24 rather than $104. That is why you need both the quota and the overage rule before comparing a subscription with pay-as-you-go pricing.

How language count changes the budget

With the same 13-minute source and the same $2.00 illustrative rate, one target language is 13 language-minutes and $26 of generation. Five target languages are 65 language-minutes and $130 of generation. The source file is unchanged, but the repeated production work is not.

Shared setup does not necessarily multiply. One cleaned transcript may support every language, although each translation still needs its own linguistic checks. Conversely, some work can grow faster than language count. A lip-sync workflow may require a separate alignment pass per language. A regulated training module may require a reviewer who understands each market. A branded voice license may be charged per voice, per language, or by usage rights rather than by minutes.

When a vendor advertises “minutes included,” check what those minutes represent. ElevenLabs’ official dubbing page shows different minutes included and extra-minute rates by plan and dubbing version, so the same source can have different effective prices depending on the selected product. ^[1] Its broader pricing page also separates plan credits from product capabilities, and displays monthly plans, credits, and enterprise custom pricing. ^[2] Record the plan name, billing period, included units, overage unit, and whether unused units roll over.

Inputs that can move the estimate

Duration and rounding. A 20-second clip may be billed as one minute, while an API can bill by characters or seconds. Do not compare those figures without normalizing the meter.

Translation scope. A translated transcript may be included, charged separately, or supplied by you. Script adaptation for timing is more work than literal translation, especially when the target language expands or contracts the spoken line.

Voice and synchronization. A basic synthetic voice, voice cloning, expressive controls, and lip-sync are different product tiers. If the source speaker’s identity or mouth movement matters, add a separate premium or request a quote.

Review. AI generation is not the same as approval. Budget pronunciation checks, terminology checks, and a short playback review per language. High-stakes content may need a full bilingual review and a second revision.

Audio mix and rights. Determine whether music and effects are preserved, ducked, replaced, or removed. Confirm commercial usage, voice-consent requirements, storage duration, and whether exports include captions or project files.

Revisions and source changes. A one-word correction may trigger a small regeneration, but a changed paragraph can affect timing and require another review in every language. Keep a revision reserve instead of treating the first render as the final cost.

For token-heavy or transcript-heavy workflows, the AI token cost calculator and LLM cost calculator can help estimate translation or orchestration calls. Those tools cover model usage; they do not automatically include dubbing-platform minutes, human review, or licensing.

A practical comparison method

To compare two providers fairly, create one row per output language. Put the exact source duration, rounding rule, included quota, overage rate, voice tier, review policy, and export policy in the row. Then add shared project costs once. This avoids the common error of multiplying a shared transcript fee by the number of languages or, in the opposite direction, multiplying a per-language dub fee only once.

Use a vendor’s official page for its current published unit and plan conditions. Use your own script length and target-language list for the workload. If a vendor provides only a custom quote, record it as quote-based rather than filling the gap with a broad online range. AWS, Google Cloud, and Azure demonstrate why: their primary speech services expose character, token, second, or video meters, each with different assumptions. ^{[3] [4] [5]}

FAQ

Is AI dubbing charged once for the whole video or once per language?

Usually, a per-language workflow repeats the dubbing output for every selected target language. If the source is 5 minutes and you request three languages, start with 15 language-minutes before applying the provider’s rounding and quota rules. Some plans bundle minutes or offer a multilingual discount, so confirm the plan’s definition of a minute instead of assuming every provider uses the same meter.

How much does it cost to dub a video into 5 languages?

There is no single defensible number without the video length, rounding rule, voice tier, and review scope. For a 10-minute source, five languages create 50 language-minutes under whole-minute billing. At an illustrative $2 per language-minute generation assumption, generation would be $100 before setup, review, storage, taxes, and optional lip-sync. Label that as a scenario, not a quote.

Should I use characters or minutes to estimate AI dubbing cost?

Use the unit the provider actually bills. Minutes are convenient for a managed dubbing product. Characters or tokens can be more accurate for a custom pipeline where you supply translated text to a speech API. A character-based estimate must still add translation, timing, audio mixing, review, and export work. Comparing a character-only TTS number with an end-to-end video-dubbing quote will understate the custom pipeline’s total cost.

Does dubbing into a new language require a new voice license?

It depends on the provider and license. A platform may bundle a cloned or synthetic voice across supported languages, while another may restrict commercial use, charge for voice cloning, or require separate approval. Treat voice rights as a separate line until the terms explicitly say they are included. Also confirm whether the voice can be used in paid ads, customer training, public social video, or client work.

Do subtitles cost less than AI dubbing?

Often, subtitles have fewer audio-production steps, but the comparison depends on translation quality, timing, formatting, accessibility, and review. A subtitle workflow does not create a new voice track or mix. If your goal is comprehension rather than an immersive localized soundtrack, price subtitles as a separate option instead of assuming the dubbing rate is the right baseline.

What is excluded from this calculator-style estimate?

The worked example excludes taxes, rush scheduling, lip-sync, music replacement, actor or voice talent fees, custom voice training, platform minimums, long-term storage, and major script rewrites. It also assumes the source transcript is usable and that review is targeted rather than a full broadcast-quality linguistic sign-off. Add those items when they are part of your delivery requirement.