AI Processing Disclosure
Which AI providers VoqText uses, for what, what they receive, and the limits of machine-generated output.
最后更新于 2026年11月6日
草案，待法律审阅
本文描述产品当前的运作方式，将在商业发布前由合格法律顾问审阅。内容可能变更。
本页目前提供英文版本。
Providers and roles
- Speech-to-text: ElevenLabs (Scribe) when configured. Receives the normalized audio track and optional language hint; returns text, timing, speaker labels and, when enabled, audio event tags.
- No language-model provider. Highlights, chapters, search and quote templates are computed by VoqText itself from your transcript: excerpts are quoted word for word, chapters are split at pauses, and nothing is summarized, paraphrased or written by AI.
- The live list of configured providers and models is shown on the Languages page and in each transcript’s provenance line.
What providers do not get
- Your account details, workspace names or other users’ content.
- Content for model training by VoqText; provider terms govern their own retention and training commitments and are linked from the subprocessor page.
Limits of the output
- Accuracy varies with audio quality, accent, language and terminology. No accuracy percentage is promised; providers publish their own per-language tiers, which are shown as theirs.
- Speaker detection can split or merge people. Edit speakers before relying on attribution.
- Automatic highlights are a selection, not a judgment of what matters most; they may miss important moments. Decisions, action items and notes in a recap are written by your team, never inferred.
- Search returns matching passages with their timestamps; where nothing matches, the product says so instead of inventing content.
- Machine output is a draft for human review, especially in legal, medical and financial contexts.