The bilingual workspace for Welsh and English broadcast, public sector, and production. Captioning, live events, document translation, and Human-Verified review - all in one place, with a Welsh-aware AI pipeline that improves with every correction.
Not another generic transcription tool. OpenAI Whisper, AssemblyAI Universal-3 Pro, and Claude AI tuned for Welsh phonetics and treigladau. Five caption styles from Broadcast to Verbatim. Professional linguist verification on demand. A complete bilingual production workflow.
Why generic AI tools don't work for Welsh content.
Major transcription tools treat Welsh as "another minority language." They miss mutations, misidentify dialects, and produce output that takes more editing than writing from scratch.
Soft, nasal, and aspirate mutations (treigladau) aren't errors - they're fundamental to Welsh. But generic AI ignores them or "corrects" them back.
Most Welsh content is bilingual - switching between Welsh and English. Existing tools force one language at a time, losing context.
Purpose-built for the unique challenges Welsh presents.
Built from the ground up for Welsh phonetics, not bolted onto an English AI.
Three tiers: Raw (normaliser only, free), Clean (Haiku AI on low-confidence segments), Broadcast-ready (full pass). Default set by plan tier.
AI splits captions at natural phrase boundaries with broadcast CPS limits.
The system tracks corrections across projects and builds a personal glossary over time.
Brand terms, dialect, formal - with voting, comments, and admin approval.
5 styles from single-word to multi-line. Choose once - it remembers your preference.
Capsiynau is part of a family of products. The same Welsh-first AI infrastructure that powers our captioning and live relay also powers Nodiadau (AI voice notes for bilingual professionals) and Llais (Welsh text-to-speech for any site).
Capsiynau is the B2B and institutional surface, built for broadcasters, production companies, and public-sector procurement. The other products are standalone consumer brands sharing the same infrastructure, the same correction loop, and the same intelligence layer.
Each one built around Welsh first. Pick the one that applies to you.
Recordings → transcripts → broadcast-ready captions. Welsh-first ASR, glossary intelligence, breath-aware segmentation.
Read moreCaptions on screen as you speak. Bilingual relay, audience page on a public URL, sub-second latency.
Read moreTranslate PDFs, Word docs and printed pages. Mutation-aware Welsh output, glossary terms enforced.
Read moreQualified Welsh linguists review and certify your captions. Broadcast standard, audit trail, signed delivery.
Read moreA bilingual editor with everything you need for broadcast-quality captions.
Edit the entire transcript as a text document - changes are auto-resynced to the original word-level timestamps.
Split segments at any point, or merge them. Adjust timings visually on the timeline.
Adjust start/end times to the millisecond - with real-time waveform preview.
Play/Pause, jump between segments, and split segments - all from the keyboard.
Clicking a segment auto-loops the audio - perfect for fine editing while listening.
See your audio visually - the waveform follows playback and click to jump to any point.
More than speech recognition - an intelligence layer that improves with every project.
Check grammar, mutations, spelling and phrasing step-by-step - review every suggestion before accepting.
Run all polish steps at once: language detection, punctuation, segmenting, Welsh cleanup, and glossary.
AI re-cuts captions at natural phrase boundaries - for smoother, more readable captions.
Corrections are tracked across projects. Repeated edits are promoted into your Welsh vocabulary -3 corrections = candidate, 10 = approved and active on the next transcription.
Manual glossary terms = hard rules (always apply). Learned terms = soft rules (context-dependent) - until promoted.
The system measures confidence based on distinct sessions where the correction appeared, not just raw count.
Your organisation's accumulated language memory - glossaries, dialects, corrections, all learning from every project.
Create multiple glossaries for brand, dialect, technical, formal, and informal terms. Private or global.
Import term lists from CSV files - map columns to "source term" and "preferred term".
Link a glossary to a folder - it auto-applies to all projects in that folder and its sub-folders.
Add comments to individual terms - discuss meaning, context, or usage.
Upvote or downvote terms - see team consensus on specific terminology.
For global glossaries: terms go through admin approval before becoming active.
Export your work in the right format for any platform.
Export captions in SRT or WebVTT format - ready for YouTube, broadcasters, and video players.
Export the transcript as plain text - with or without speaker labels and timestamps.
Export all languages, Welsh only, or English only - perfect for single-language subtitles.
Add translations in any language - export alongside the original transcript.
Upload directly from an HLS / .m3u8 stream URL - no need to download the file first.
MP3, WAV, MP4, MOV, PDF, DOCX supported on upload - audio extraction handled for videos.
For events, broadcasts, council meetings and live-streamed lectures - bilingual captions appearing alongside the speech, not after it.
Sub-second latency for English speech. Captions land on the audience screen before the next sentence starts.
Welsh-aware live transcription with mutation cleanup and dialect support running on the same pipeline.
Project the live transcript to a screen at your venue, or hand the audience a QR for the bilingual feed on their phones.
Built for the bilingual delivery duties councils, Senedd committees, and Welsh public sector organisations are required to meet.
Your team's place names, role titles, and brand terms - and BydTermCymru's 76,000+ terms - all corrected on the fly.
Live transcripts save as projects you can open in the editor for a tidy broadcast-grade version after the event.
PDF, Word, ODT, and plain text. Bilingual translation that preserves formatting and respects the conventions of formal Welsh writing.
Upload .pdf, .docx, .odt, .txt. Receive the translated version in your choice of format, plus side-by-side bilingual view.
Your organisation's term canon applies to every document. Place names, role titles, and policy terminology stay consistent.
Soft, nasal, and aspirate mutations applied correctly - the difference between machine-translated Welsh and Welsh that reads naturally.
Send the AI translation to a Human-Verified linguist for a professional pass. Returns to you with tracked changes you can accept or amend.
Designed for council reports, NHS Wales materials, education resources, and any document that needs the same authority in both languages.
Headings, lists, and tables hold their structure. Translation lands where it belongs without breaking your layout.
Qualified Welsh linguists review, correct and certify your captions. Used by broadcasters, the public sector and creators where accuracy is non-negotiable.
Submit duration, subject, language direction and deadline. Admin sets a price within hours.
Once accepted, a Welsh-fluent translator from the approved pool picks up your project in their portal.
Every edit is logged in an audit trail. Broadcast-standard segmentation and mutations checked end-to-end.
SRT, VTT and EBU-TT delivered via signed download links. PDF invoice attached. 24h link expiry.
Everything you need to go from audio file to professional captions.
Purpose-built for Welsh phonetics, soft/nasal/aspirate mutations (treigladau), and dialect variation - not a generic AI.
Every word timestamped to the millisecond. Perfect sync for captions, subtitles, and waveform editing.
Identify multiple speakers with colour-coded labels in the editor. AI-assisted with human confirmation.
Auto-detect Welsh and English per segment - perfect for bilingual content.
Add punctuation, capitalisation, and sentence structure using AI - for both Welsh and English.
5 caption styles: single-word, single-short, single-long, multi-short, and multi-long - with broadcast CPS limits.
Organise projects into folders and sub-folders. Drag and drop to reorganise. Glossaries linked to folders.
Upload an original script - the system aligns words to timestamps, correcting wording without losing sync.
Save snapshots of your transcript at any point. Restore previous versions easily.
Embed Capsiynau in internal tools, workflows, and AI agents. Turn content into language intelligence.
Convert audio/video to structured captions
AI-assisted summaries, chapters & JSON-LD metadata
Apply scripts, glossaries & AI improvements
Flag sensitive topics & brand mentions
RESTful API with webhooks, async jobs, and SDKs for your language.
Optimized for streaming, batch processing, and real-time captions.
Track usage, errors, and performance with detailed analytics.
Use the official Node SDK to submit transcription jobs, poll status, and fetch subtitles in a few lines of code.
npm install @capsiynau/sdk-transcribe
Sign in to view full API documentation
Built by producers, for producers. Not a tech project - a production tool.
Built by people who work in Welsh broadcast production - not by tech developers distant from the industry.
Battle-tested in real production workflows with Welsh broadcasters and content creators.
Welsh isn't an afterthought. Every design decision - from the AI engine to the UI - starts with Welsh.
We build in the open. Features are driven by real feedback from Welsh producers and broadcasters.