Research-backed calculator guide
Subtitle Translation Cost per Finished Minute: Include Timing, Formats and QA
Estimate subtitle translation cost per finished minute by separating transcription, translation, timing, QA, file delivery and optional burn-in.
A subtitle quote is only comparable when “per minute” means the same thing. A machine transcription API may meter audio seconds, a translator may price source words, and a localization vendor may quote by finished video minute. Those are different units. The finished-minute number should include the work needed to deliver an accepted, timed subtitle file—or a burned-in video if that is what the brief requires.
Quick answer
For a practical estimate, use:
Cost per finished minute = (shared source preparation + language work + QA + delivery + fixed fees) ÷ accepted video minutes.
For one target language, a useful line-item model is:
Total = transcription or transcript cleanup + translation + timing/spotting + linguistic QA + file formatting/export + project fee + optional burn-in.
Do not add every line automatically. If the source transcript is already accurate and time-coded, transcription may be zero, but review and timing validation still need an assumption. If you need two target languages, source preparation can remain shared while translation, language QA and language-specific exports generally repeat. If you need open captions, add the burn-in or video-rendering step separately from the switchable SRT or VTT deliverable.
As an order-of-magnitude planning model, many professional projects land far above the raw API meter because people still have to translate for meaning, condense text for screen space, check cues, correct names and export the requested format. A vendor quote is the source of truth for a purchase. The figures below are a transparent estimate, not a universal market rate.
What “finished minute” should include
Define the denominator before comparing translated subtitles cost per minute. A finished minute is one minute of source runtime that has passed the agreed acceptance check. It is not one minute of translator time, one minute of audio successfully transcribed, or one minute of a preview render.
Start by deciding whether the supplied asset is a video, clean audio, a transcript, or an already timed subtitle file. A video with no transcript needs source transcription or transcript creation. A clean transcript can reduce source-language typing, but it does not automatically contain cue boundaries, reading-speed decisions, speaker changes or subtitle-safe line breaks. A properly time-coded SRT or VTT can remove substantial setup work, but it still needs a source review and target-language validation.
The output also matters. Amazon Transcribe documents SRT and WebVTT subtitle output and notes that a subtitle job can produce both the subtitle file and a regular transcript file.[2] That is a format capability, not proof that the result is translation-ready or client-ready. A translated subtitle track needs its own language work and quality review. If the deliverable is a video with captions permanently visible, the export adds rendering time and may require checks for safe areas, contrast and collisions with existing graphics.
Keep three outputs separate in the quote:
- Timed subtitle file: a switchable SRT, VTT, STL, TTML or platform-specific file.
- Preview or approval asset: a low-resolution video or review link used to catch timing and visual issues.
- Burned-in master: a rendered video with captions embedded in the picture.
They may share the same subtitle cues, but they are not the same deliverable.
The cost components
1. Source transcription and cleanup
This is the cost of turning speech into a usable source-language transcript. It may include diarization, speaker labels, punctuation, terminology correction and removal of false starts. Treat API pricing as a production input, not as the finished subtitle price. AWS says Amazon Transcribe is pay-as-you-go based on seconds of audio, billed in one-second increments with no minimum applied; its displayed examples are region- and volume-specific.[1] Google Speech-to-Text likewise describes usage in processed audio seconds, and its pricing page says multiple channels are billed separately.[3]
For a stereo or multichannel recording, confirm whether the provider bills the sum of channels or only total runtime. A one-minute file can therefore represent different billable usage depending on the service and request. Add storage, orchestration or retry assumptions only when they are actually part of your pipeline.
2. Translation
Translation is usually the largest language-dependent line. It is not just replacing words. Subtitle translation must preserve meaning and tone while fitting cue duration, line length and reading speed. Charge it per source minute only when the content density is reasonably stable; otherwise use source words or characters and convert that estimate to a finished-minute view for comparison.
For dense training, legal, medical or technical material, use a higher planning rate or a separate complexity multiplier. A 60-minute runtime does not tell you whether there are 2,000 or 12,000 source words. Ask for a sample or transcript before turning a per-minute estimate into a quote.
3. Timing, spotting and condensation
Spotting places in and out timecodes around readable chunks. It also includes deciding where to split lines, how to handle speaker turns and whether translated text must be condensed. Even when a source subtitle file exists, target-language expansion or contraction can make cue timing and line breaks invalid. Put this work on its own line instead of hiding it inside “translation.”
4. Linguistic QA and subtitle review
A review pass checks omissions, mistranslations, names, terminology, punctuation, line breaks and timing against the video. A second linguist is different from an automated spellcheck. If accessibility, regulatory, broadcast or brand requirements apply, identify the rule set before pricing the review. The subtitle review cost per video minute should include the review scope, number of passes and who resolves disputed changes.
5. Formats, delivery and burn-in
Exporting an SRT or VTT may be a small per-file task, but multiple platforms can require different conventions. Confirm numbering, encoding, frame rate, timebase, cue overlap and maximum line length. AWS warns that subtitle numbering defaults can affect compatibility, which is a reminder to validate the target platform rather than assuming every SRT is interchangeable.[2]
Burn-in is an optional production step. It can be priced per video minute, per output, or as an editing/rendering package. It is not equivalent to delivering a subtitle file. If you deliver both, list both. Include a visual inspection of the encoded result when text can be obscured by existing lower thirds, logos or on-screen captions.
6. Fixed project fees and minimums
Small jobs are distorted by fixed costs. Intake, language setup, terminology preparation, file handling and final delivery may take nearly the same effort for a five-minute clip as for a longer episode. Allocate a fixed project fee across the finished minutes for comparison, but preserve the original fixed amount in the quote so the client can see the minimum-fee effect.
Worked cost example
Assume a 30-minute English training video and one Spanish subtitle track. The client supplies a clean video but no reliable transcript. The estimate below uses a hypothetical planning rate for human work. The $0.006 per audio minute transcription input is an illustrative AWS-style batch assumption from the documented US East example, not a guaranteed quote for this project.[1] The human rates are assumptions chosen to show the calculation, not vendor prices.
| Step | Assumption | Input | Output |
|---|---|---|---|
| Source transcription | Automated batch transcription at $0.006/audio minute | 30 min × $0.006 | $0.18 |
| Translation | Planning allowance of $12 per finished minute | 30 min × $12 | $360.00 |
| Timing and spotting | Planning allowance of $3 per finished minute | 30 min × $3 | $90.00 |
| Linguistic QA | One review pass at $4 per finished minute | 30 min × $4 | $120.00 |
| File formatting/export | One target-language subtitle package at $1 per minute | 30 min × $1 | $30.00 |
| Project setup and delivery | Fixed fee allocated to this job | 1 × $30 | $30.00 |
| Finished subtitle file total | Sum of required lines | — | $630.18 |
| Cost per finished minute | $630.18 ÷ 30 accepted minutes | — | $21.01/min |
| Optional burn-in | Additional render/inspection at $5 per minute | 30 min × $5 | $150.00 |
| File plus burned-in video | $630.18 + $150 | — | $780.18 total / $26.01 per finished minute |
This example makes two important assumptions visible. First, the low machine meter is not the price of the deliverable. Second, burn-in changes the scope. If the client supplies a verified, time-coded source SRT, remove or reduce the source transcription line, but do not silently remove translation, target-language timing validation or QA. If the project has three target languages, the $30 setup fee might remain shared while the translation, review and export rows repeat three times. Decide whether a vendor’s minimum fee applies per language, per file or per project before comparing quotes.
For an API-heavy workflow, the same model still works. OpenAI’s published API table, for example, lists some transcription and live-translation prices per minute, while other models are priced per token.[4] Put the actual service meter in the source-preparation or machine-assist row, then add the human and delivery rows that turn the output into a finished subtitle asset. If a model translates text but does not preserve timing, its API cost belongs in machine translation, not in the entire finished-minute estimate.
Run your scenario
Turn this framework into an editable estimate.
Use dated, source-linked assumptions and see cost per request, successful task and month.
How to estimate multiple languages
Separate shared and repeating work. A simple two-language model is:
Total = shared source work + (translation + target timing + target QA + target export) × language count + fixed project fee.
That formula is more useful than multiplying the entire one-language quote. Source transcription, terminology extraction and ingest may happen once. Translation and linguistic review normally happen once per target language. Burn-in may also repeat because each language creates a different video master. If the client only needs switchable subtitle files, do not price a burned-in master by default.
Language count is not the only multiplier. Right-to-left scripts, complex typography, vertical video safe areas, speaker labels, SDH sound descriptions and platform-specific constraints can add work. Put those as explicit adjustments with a reason. Avoid pretending that a generic “language multiplier” captures every difference.
Common mistakes that inflate or hide the number
The first mistake is dividing by the wrong denominator. A project with a 28-minute accepted runtime and a $700 invoice costs $25 per finished minute, not $23.33 because the source file was nominally 30 minutes. State whether intros, credits, silence and rejected sections are included.
The second mistake is mixing a vendor quote with an API estimate. A published service price may bundle human review, file delivery and turnaround. An API meter usually does not. Date any changing provider price, record region and model, and keep taxes, storage, rush fees and platform fees separate.
The third mistake is treating a raw transcript as a timed subtitle file. Transcription answers “what was said.” Timing and subtitle layout answer “when should this readable piece appear, for how long, and where should it break?” Those are different acceptance tests.
The fourth mistake is counting QA as optional when the brief promises a finished deliverable. You can offer a machine-only draft, but label it as a draft and give it a different acceptance level. Do not compare that number directly with a human-reviewed subtitle quote.
For tokenized translation or cleanup stages, the AI token cost calculator can help estimate model input and output separately. For a broader workflow with several model calls, use the LLM cost calculator. If you are comparing provider meters, the Model pricing comparison guide explains why a unit price needs a common workload before it becomes a fair comparison.
FAQ
What is the difference between translated subtitles cost per minute and cost per finished minute?
Translated subtitles cost per minute may mean only the language conversion. Cost per finished minute divides every required project cost by accepted runtime, including source preparation, timing, review, formatting, fixed fees and any specified burn-in. Always ask which work the quoted minute includes.
Does an existing transcript reduce subtitle translation cost?
It can reduce transcription, but only if it is accurate and usable. A plain transcript still needs cueing, condensation, timing and review. A verified, time-coded source subtitle file can remove more setup work, but target-language cues still need validation. Treat the supplied asset as an input assumption, not as an automatic discount.
Should I price SRT, VTT and burned-in captions the same way?
No. SRT and VTT are timed-text files with different formatting and compatibility checks. Burned-in captions require a video render and visual inspection, and they cannot be switched off. Quote file delivery and burn-in as separate outputs even when both use the same translated cues.
How do I estimate a multilingual caption cost calculator result?
List shared source work once, then repeat language-dependent translation, timing validation, QA and export for each target language. Add language-specific typography, SDH, RTL, rush or platform requirements as explicit adjustments. Finally divide the combined total by accepted source minutes, not by the number of files.
Are transcription API prices a good proxy for a subtitle quote?
They are useful for the machine portion only. AWS and Google document audio-time metering, and OpenAI lists a mixture of per-minute and token-based audio prices.[1] [3] [4] None of those meters, by themselves, includes the complete human-reviewed translation and delivery workflow. Use the API rate as one row in the finished-minute model.
What should a subtitle review cost per video minute include?
Define the review pass. It should state whether the reviewer checks meaning, terminology, omissions, cue timing, line breaks, reading speed, speaker labels, sound descriptions, format validity and the rendered video. One proofreading pass on text is not the same service as full audio-and-picture QC.
Limitations and assumptions
This model is a budgeting framework, not a universal price list. Human rates vary by language pair, content density, specialization, turnaround, country, minimum fee and whether the quote is per source minute, target word, file or hour. Provider prices change; the official API pages cited here were checked on 2026-11-02, and the displayed rates may depend on region, model, tier, account and volume. Taxes, storage, editing, rush work, accessibility certification and platform submission are excluded unless a row explicitly includes them.
The benchmark behind this guide used English, United States as the requested locale. The search tool did not expose a reliable country-specific sponsored-versus-organic label or a browser location control, so the SERP observation records that limitation. Search coverage describes the pages returned on the checked date; it does not predict rankings, traffic, indexing, conversions or revenue.
Primary sources checked
Pricing and product behavior can change. These sources support the time-sensitive claims in this guide; verify current provider terms before committing spend.