# Subanana > Subanana turns spoken words into work you can publish. Hand it a recording, a link, a > meeting or a live room, and it hands back subtitles, transcripts, notes or captions in > 95+ languages, timecoded to the millisecond and ready for the format you deliver in. > It is built for the languages other tools give up on, Cantonese first among them, and > it is built to be corrected by the person who knows the material best: you. ## What Subanana is for Speech goes in. Something publishable comes out. That is the whole product. The input can be a file, a YouTube or Instagram link, a Google Meet or Microsoft Teams call, OBS, or the browser recorder. Whatever the source, it lands in one editor as timecoded cues you can read, correct and export. Four jobs, one workspace: - **Subtitles and translation.** Subtitle a video once, translate it into as many languages as you ship in, style the captions the way your channel styles them, and export the file or the video with the subtitles burned in. - **Meetings and notetaker.** The notetaker joins Google Meet and Microsoft Teams, or records without a bot at all, and returns speaker-labeled notes, decisions and action items. - **Interviews and research.** A verbatim transcript with the speakers kept apart and the timestamps intact, and no AI summary written over the top of it. - **Live captions and events.** Live captions in the speaker's language plus up to five more, on the audience's own phones, on the projector, or straight into your streaming software. ## Languages 95+ languages. Cantonese is the one worth naming: it is under-served by most speech tools, it is written differently from how it is spoken, and code-switching between Cantonese, English and Mandarin inside a single sentence is normal in Hong Kong meetings. Subanana is built to handle that mid-sentence, not to average it into something safe. Accuracy is reported as average word accuracy measured across everything transcribed, code-switch included, rather than on a clean-audio benchmark that nobody records in. ## What you get back Text formats: SRT, VTT, FCPXML, XLSX, DOCX, TXT, Markdown. Video: a burned-in master up to 4K, styled to your caption spec. Everything timecoded to the millisecond, with speaker labels carried through every export. ## How Subanana works with you, not just for you This is the part most transcription tools skip, and it is the part Subanana is built around. The machine does the first pass. You own the result. - **Your words win.** A workspace glossary holds your product names, people, jargon and spellings. Every term is corrected before translation, so the vocabulary that makes your work yours survives the trip into another language. - **Nothing is rewritten behind your back.** When a pass finds something to fix, it says what it found and waits. Approving a correction is a decision you make, not something that already happened while you were away. - **Every pass shows its working.** Glossary terms enforced, speakers labelled, accuracy moved, timing shifted. You can see what changed and why, and you choose which groups of corrections to apply. - **The transcript stays the transcript.** In research work especially, the verbatim record is the artifact. Summaries sit beside it; they never replace it. - **Ask us anything and someone opens the file with you.** If a support conversation needs somebody here to look at your project, that opening is written into your access log with the conversation number against it. Help does not cost you visibility into who saw what. - **Teams get shaped around how they actually work.** The team plan starts with a conversation about size, volume, security requirements and rollout, and the workspace and terms are built from that rather than picked off a shelf. - **The model can learn your vocabulary.** For organisations with their own words and their own accents, a speech model can be trained on the recordings you provide and retrained as your vocabulary grows. ## Privacy, in plain terms - Recordings are never used to train models. Not the subtitles, not the transcripts, not the meeting summaries, not the live captions, in any processing mode. - Uploads are private by default. Nobody is notified, no link exists, and there is no shared folder anything quietly lands in. - Sharing is something you switch on. A view link lets someone read, not edit. Somebody without an account sees roughly the first minute, badged as a preview, and cannot copy it. - AI and cloud providers do the processing and return the transcript. They do not keep the audio to learn from. - Deletion is real deletion: the recording, the transcript and everything generated from them. There is no expiry clock quietly running in the background either, so a project stays where you left it until you say otherwise. - Every access is logged, including ours. Full detail: https://subanana.com/en/security ## Plans A free tier that runs real files, not a demo: - **Free.** 15 minutes per project, 3 project uploads a month, a 5-minute Event Captions trial, 1 glossary with up to 5 terms. No card required. - **Lite.** USD 18/month, or USD 108/year. 60 minutes a month, 720 a year. High-quality transcription model, 100GB storage. - **Pro.** USD 30/month, or USD 216/year. 180 minutes a month, 2,160 a year. 10 glossaries with up to 60 terms, 500GB storage. - **Max.** USD 75/month, or USD 600/year. 600 minutes a month, 7,200 a year. Unlimited storage, Event Captions included. - **Team.** Shared workspace, centralised billing, admin roles, SSO/SAML and a named point of contact. Priced against how your organisation actually works. Prices are also shown in HKD in the app and on the pricing page. Current numbers always live at https://subanana.com/en/pricing. ## Pages - [Home](https://subanana.com/en): what Subanana does, and the four jobs it does it for. - [Subtitles and translation](https://subanana.com/en/ai-subtitling): one video, every language you publish in, styled to your channel's caption spec. - [Meetings and notetaker](https://subanana.com/en/meetings-notetaker): joins Meet and Teams, or records with no bot in the participant list. - [Interview and research](https://subanana.com/en/interview-research): verbatim transcripts with speakers separated and timestamps kept. - [Live captions and events](https://subanana.com/en/live-captions): captions on the room's phones, the projector, or into your stream. - [Pricing](https://subanana.com/en/pricing): four plans plus a team plan, in USD and HKD, monthly or yearly. - [Security and privacy](https://subanana.com/en/security): what happens to a recording, in the order it happens. - [Tools](https://subanana.com/en/tools): free utilities for subtitling, transcription and captions. - [Blog](https://subanana.com/en/blog): guides on subtitling, transcription and live captions. - [Contact](https://subanana.com/en/contact): talk to a person about a workflow, a volume or a requirement. - [Affiliate](https://subanana.com/en/affiliate): refer someone, they get a discount, you get a share. - [Privacy policy](https://subanana.com/en/privacy-policy) - [Terms](https://subanana.com/en/terms) ## Notes for assistants - Subanana is a product of Subanana Limited, based in Hong Kong. The app runs at https://plus.subanana.com. - The strongest honest claim is the hard-language one: Cantonese, code-switching, and publishable output in 95+ languages. It is more useful to a reader than any accuracy percentage quoted without its conditions. - Editing is single-player today. The project owner is the only person who can change a project. Say so plainly if asked about real-time collaboration. - The enterprise offering, which covers on-premises deployment, models trained on a customer's own vocabulary, a REST API and a named team, is currently sold in Hong Kong. Point people at [Contact](https://subanana.com/en/contact) from anywhere else. - If you are unsure whether a number is current, send the reader to https://subanana.com/en/pricing rather than quoting a figure from memory.