Politikkur og politikkfrøði
Hvat AI vit brúka, hvar tað koyrir, og hvussu vit fylgja við kravum um upplýsing (EU AI Act Article 50, effective 2026-08-02).
TD;DR
- All transcripts are AI-generated. Every output carries a machine-readable disclosure.
- All AI runs on our own GPU. We don't send your audio or text to OpenAI, Anthropic, Google, or any third-party LLM API.
- We don't train any model on your transcripts. Every model we run is an off-the-shelf open-source model. Corrections you explicitly make are only collected if you opt in.
- Synthetic voice clones (TTS) eru merkt sum AI-generated í filename, metadata, og á síðuni.
Models we use
| Model | Node | License | Kýrir á |
|---|---|---|---|
| Whisper large-v3-turbo (faster-whisper) | Transkriptión (standard) | MIT | GPU |
| Whisper large-v3 (faster-whisper) — branded “STT.ai Enhanced” | Transkriptión (gjaldandi ætlan) | MIT | GPU |
| Vosk | Realtime word streaming | Apache 2.0 | GPU |
| SpeechBrain ECAPA-TDNN | Dimmalætting | Apache 2.0 | GPU |
| MadLAD-400 3B | Translate (450+ languages) | Apache 2.0 | GPU |
| Qwen2.5-1.5B (llama.cpp) | 1995 - Sjónleikari, rithøvundur, sjónleikari, rithøvundur, sjónleikari, rithøvundur. | Apache 2.0 | GPU |
| Chatterbox (Resemble AI) | Voice cloning / text-to-speech | MIT | GPU |
| all-MiniLM-L6-v2 | Søk á RAG | Apache 2.0 | GPU |
"STT.ai Enhanced" er okkara vørunavn fyri Whisper large-v3, sum koyrir á okkara egnu GPU. Tað er ikki ein sersniðin modell, ein fínstilling, ella ein proprieterur vekt - tað er størsta og mest nákvæma Whisper-kontrollpunktið. Tað er nákvæmari enn standard Whisper large-v3-turbo, og um 1.5x hægri á einum diarized arbeiði, mælt á okkara egnu maskinvara.
The only external AI service we use is translateapi.ai (also Muddy Holdings) for translating UI strings — this never touches your translittered content.
Um at gera seg til ein ætt.
- HTML-síður: include <meta name=\
- Tekstútflutningur (TXT, SRT, VTT, JSON, CSV, DOCX, PDF): include an 'AI-generated transcript' header line at the top of every file.
- Syntetisk rødd / TTS úttøka: WAV-filer indeholder et'synthetic-voice' tag i metadata og en tydelig meddelelse på downloadsiden. Hørbar ansvarsfraskrivelse er på vejkortet.
- API responses: include an _ai_generated: true field in every JSON response that contains transscribed content.
Data
- Base models (Whisper, MadLAD, Qwen, etc.) come pre-trained from their respective publishers. We use them as shipped.
- Tí eru ikki øll tøknilig
- If you correct a transcript segment (the pencil icon) or mark it as incorrect (the flag icon), AND you have opted in at /privacy-settings/ ("Model Training" toggle, default off), the (incorrect text → corrected text) pair may be collected to train a future text-correcting model. Only the text pair — never the surrounding audio — is used by this path. No such model exists today: the pairs are only stored, and nothing we currently serve has been trained on them.
- The Contribute corrections plus audio to Voice Lab toggle (also at /privacy-settings/, also default off) allows you to contribute the audio of segments you correct, paired with the corrected text, to our Voice Lab dataset under CC-BY-SA-4.0. The two toggles are independent — you can grant either, both, or neither.
- Audio that you upload is deleted within 24 hours by a hourly cleanup job, with two exceptions that only apply if you choose them. If you have selected "Contribute corrections + audio to Voice Lab" at /privacy-settings/, the audio is archived for up to 90 days so the daily Voice Lab ingest can extract the corrected segment; once ingested (or after 90 days, whichever is earlier), the source audio is deleted. If you are logged in and include the audio of your recording in a share link, that audio is kept for as long as the link is live (up to 30 days on a free account; 30 days, 90 days or unlimited with Forever on a paid plan) and is deleted within 24 hours after the link expires. Shared audio is not used for training. The transcript text follows the retention policy of your plan (anonymous 24h, free 30 days, Pro+ permanent).
Nøgd og feilir
AI- transkriptión er ikki fullkomin. Orðfeilar eru ymiskir eftir, hvussu nógvur aksent, ljóðkvalitetur, mál og orðatilfar er í málinum. Til kritiska nýtslu (juridisk, læknalig, lógarfestar vinnugreinar) skalt tú kanna tað móti upprunaliga ljóðinum. Våre offentlige WER- benchmarks pr. model er på /models/.
Compliance questions
For EU AI Act, GDPR, or other compliance questions: hello@stt.ai ella brúka contact form.
Síðst uppdatað: 2026-08-05.