Dub a 60-second clip into Hindi: Sume step costs before translation
A 60-second dub on Sume is four steps. Detach audio $0.01, STT $0.01, translate with your own tool, then TTS in Hindi: about $0.0628 for 900 characters.

The steps and the cost
Sume has no single dubbing endpoint, so a dub is a chain: detach the audio ($0.01), transcribe it ($0.01 for one minute), translate the text yourself, then synthesise it in Hindi with TTS. For a 60-second clip and a Hindi script of about 900 characters, the Sume steps total $0.01 + $0.01 + 900 x $0.0475 / 1,000 = $0.01 + $0.01 + $0.04275 = $0.06275, before translation and any editing time.
The 900 characters is an assumption for planning. Count your real translated text and use the exact length.
Step table
Each price comes from the Sume docs or the repository's golden pricing file.
| Step | Route | Cost |
|---|---|---|
| 1. Detach the audio | POST /v1/audio-detach | $0.01 |
| 2. Transcribe | POST /v1/stt-1.0/transcribe, 60 s | $0.01 |
| 3. Translate | Your own tool or model | Not a Sume route here |
| 4. Hindi voice | POST /v1/tts-1.0/generate, language hi, 900 characters | $0.04275 |
| Total (Sume steps) | $0.06275 |
How each step works
Detach the audio from a media.sume.com video as a 16 kHz mono file, or an mp3, then pass its URL to STT with duration_seconds: 60. Use words[] or sentence segmentation to see where each line starts, so your translation can follow the same rhythm.
Translate sentence by sentence, keeping each line close to the original length. Hindi often runs longer than English, so the dub may exceed the slot. Fix that by trimming the translation first, and use generation_config.speed (0.6 to 1.5) as a last resort, since a fast voice sounds hurried.
For the final voice, set language to hi and pick a voice that speaks Hindi. Join per-sentence audio files in order with timeline audio concat, then place the audio back on the video.
Honest limits
This keeps the original timing only as well as your translation allows. It does not change lip movements; if the speaker's face is on screen, the mouth will not match the Hindi. A voiceover over a screen recording or b-roll is the easiest case. Check the result with a Hindi speaker before you publish.
- Keep the original audio file in case you redo the translation.
- Number your sentences so audio files stay in order.
- Keep the translation in a separate file you can edit.
- Never resubmit a paid job that looks slow; poll its status.
Scaling the estimate
The same arithmetic applies to longer clips. A 5-minute video with 4,500 translated characters would cost one detach ($0.01), 5 minutes of STT ($0.05) and 4.5 x $0.0475 = $0.21375 of voice, which is $0.27375 in total. Keep the sources under the limits: audio detach reads up to 1,800 seconds, and each STT job takes at most 600 seconds. Long dubs are best built scene by scene, so one bad translation costs you one scene of voice, not the whole video, and you can send the scenes to a reviewer in parallel.
Sources
Related posts
More in Use cases
- eBay AI listing photos: the seller owns accuracy
eBay's User Agreement makes sellers responsible for listing accuracy, including generative AI content. Generate 1:1 stills from your real photo and check them.
- E-commerce product video cost per SKU: cutout plus a 5-second clip
A cutout at $0.0225 plus a 5-second 720p Seedance 2.5 clip is $2.9115 per SKU on Sume. For 100 SKUs that is $291.15. Arithmetic inside.
- ElevenLabs free-plan music: the ElevenMusic credit rule for ads
Free ElevenLabs users may use tracks commercially only with an ElevenMusic credit; Pro gets rights with the plan. What that means for a paid video ad.
- Estate sale preview video from mixed portrait and landscape photos
Phone photos of estate sale lots come in both orientations. Timeline's fit modes (cover, contain, stretch, blur) decide whether a lot is cropped or kept whole.
Written by Sume