AI Transparency Statement
Datganiad Tryloywder AI
Last updated: March 2026
Overview
Capsiynau uses artificial intelligence to provide transcription, captioning and language analysis services. This page explains how AI is used, its limitations, and how your data is protected when it is processed by AI systems.
How AI Is Used
| Function | Description |
|---|---|
| Speech transcription | Converting audio to text using speech recognition models |
| Caption segmentation | Dividing transcripts into timed caption segments |
| Welsh normalisation | Correcting Welsh mutations, diacritics and contractions |
| LLM refinement | Improving grammar, punctuation and readability |
| Translation | Translating Welsh captions to English or other languages |
| Terminology suggestions | Identifying domain-specific terms for glossary review |
AI Providers
| Provider | Purpose | Data Not Used for Training |
|---|---|---|
| OpenAI (Whisper, GPT-4o) | Transcription and translation | ✅ Via enterprise API |
| AssemblyAI | Transcription | ✅ Model training opt-out enabled |
| Anthropic (Claude) | Caption refinement and analysis | ✅ Via API — not used for training |
| Google Cloud (Chirp 2) | Welsh transcription | ✅ Enterprise configuration |
Your Content Is Not Used to Train AI Models
Capsiynau does not train its own AI models using your content. Where third-party AI providers are used, Capsiynau configures services to ensure submitted data is not used to train models wherever technically possible.
AI Limitations
AI-generated transcripts may occasionally contain:
- Transcription inaccuracies, particularly for proper nouns, place names or technical terms
- Punctuation errors
- Speaker misidentification in multi-speaker recordings
- Errors with Welsh mutations (treigladau) or dialect variations
For broadcast or professional publication, all AI-generated transcripts should be reviewed and edited by a qualified human operator before use.
Human Oversight
Capsiynau is designed to support human editorial workflows, not replace them. The platform provides:
- Confidence scores to highlight uncertain segments
- A full editing interface for reviewing and correcting transcripts
- A Human-Verified translation service for critical content
- Version history so changes can be reviewed and reversed
Welsh Language Commitment
Capsiynau is built specifically for Welsh and English broadcast production. Our Welsh language AI features include:
- Welsh vocabulary boost (652+ broadcast terms)
- Welsh mutation normalisation
- Welsh dialect support (North Wales, South Wales, Broadcast)
- Welsh-first prompting for all transcription engines