Fact · Voice privacy and security
HoldSpeak's security document says a pending browser voice capture is kept only while transcription has not confirmed its text, and that a successful transcription deletes it. The document gives no number of days.
Quote found word for word on the live source, last checked October 25, 2026
In the source's own words
“One bounded WAV per scope, kept only while transcription has not confirmed text. A successful transcription deletes it.”
HoldSpeak SECURITY.md: Security and privacy (GitHub repository), karolswdev, GitHub (karolswdev/HoldSpeak), 2026, Section 1, Data classes, Web pending voice capture row.
Open the original- What it measures
- The row covers the recovery copy of a voice capture kept in the browser's IndexedDB under holdspeak-voice-recovery. The copy is kept for a capture that has not yet produced confirmed text.
- What it does not show
- The rule covers the browser's recovery copy, not the text transcription returns. The document does not say how long a capture is kept if transcription never succeeds, so the row has no number of days. It is the maker's own description and has not been audited.
In the data
A row of How Long Voice Assistants Keep Your Recordings, as HoldSpeak, web pending voice capture: kept only while transcription has not confirmed text. A successful transcription deletes it days.
- Service
- HoldSpeak
- What is kept
- Audio recordings
- When
- Other
- As of
- 2026-10