Subprocessors
Last updated · October 20, 2026
These are the third parties Dubsmith engages to run the service.Dubsmith never trains AI models on your content. These providers process it under their own terms, which may allow them to use it, including to train or improve their AI models. This page is the list referenced by the Privacy Policy and the Data Processing Addendum.
Platform and business operations
These providers handle your account, the application itself, billing, and email.
| Provider | What it does | What it receives | Where |
|---|---|---|---|
| Convex | Application database, authentication records, review state | Account records, project and job metadata, transcripts and translations | United States |
| Cloudflare | Website and application hosting (Workers), media object storage (R2) | Uploaded source media, rendered deliverables, request metadata | Global edge network; object storage in the configured R2 location |
| Amazon Web Services | Media processing pipeline — API compute, container workloads, transient working storage, job queues | Working copies of media and derived artefacts for the duration of a job | United States (US East, N. Virginia) |
| Stripe | Payment processing, subscription and usage billing | Billing contact details, transaction and metered-usage records. Card details go to Stripe directly and never reach our servers. | United States, Ireland |
| Resend | Transactional email (sign-in, password reset, job notifications) | Email address, name, message content | United States |
| Crisp | Live support chat on the website and in the app | Chat messages and files you send, your IP address, browser type and the page you are on; when signed in, your name, email address and account ID | European Union (Netherlands, Germany) |
| PostHog | Product analytics — pageviews and named product events, cookieless (privacy policy §5) | Pseudonymous account ID, pages viewed, product events with plan, language and output type, browser type, IP address (for approximate location), app errors. No media, transcripts, names or email addresses. | United States |
Media processing
These providers perform steps of the localization pipeline and see the media itself. Each receives only what its step requires, for the duration of that step.
| Provider | What it does | What it receives | Where |
|---|---|---|---|
| Google Cloud (Gemini on Vertex AI, Video Intelligence, Cloud Storage) | Transcription, translation, emotion and on-screen-text style analysis; frame-accurate on-screen text detection; short-lived staging of the media those steps read | The dialogue audio track, the source video, still frames and crops of on-screen text, and the transcripts and translations derived from them | Global — Gemini requests are served from any Google Cloud region |
| AudioShake | Source separation — splitting the soundtrack into dialogue and music-and-effects tracks | The audio track of the source media | United States |
| Mixfont | Font identification for localized on-screen text | Small image crops of on-screen text taken from video frames, fetched through short-lived links | United States |
| Alibaba Cloud (Qwen) | Translation (alternate provider, used only when selected in place of Gemini) | Transcripts and on-screen text to be translated. No audio or video. | Singapore (storage); processing on Alibaba Cloud nodes outside mainland China |
| Fish Audio | Speech synthesis and voice cloning (primary voice provider) | Translated text segments and reference audio of each speaker, cut from the source, used to clone their voice | United States |
| MiniMax | Speech synthesis and voice cloning (fallback provider for languages other than English) | Translated text segments and reference audio of each speaker, cut from the source, used to clone their voice | Singapore, United States |
| ElevenLabs | Speech synthesis and voice cloning (fallback provider for English) | Translated text segments and reference audio of each speaker, cut from the source, used to clone their voice | United States |
| RunPod | On-screen text removal (video inpainting); on-screen text detection when selected in place of Google Video Intelligence | Video segments and the removal masks derived from them; for text detection, the source video | Data centres worldwide, including the United States |
Voice providers. Which speech provider handles a job depends on the target language, with a second provider used only if the first is unavailable, so a given job will normally involve one of them rather than all three. Dubs keep each speaker’s voice, so the provider receives reference audio of the speaker cut from your source — see the consent requirements in the Acceptable Use Policy.
Changes to this list
Before a new subprocessor begins processing customer personal data, we update this page and give business customers at least 15 days’ notice by email to the address on the account. Customers under the Data Processing Addendum may object on reasonable data-protection grounds within that window; the process and remedy are set out there.
To be notified of changes, or to ask about a specific provider, email contact@dubsmith.io.