MarkVox Voice Data Policy

Last updated: 2026-06-21 Last checked: 2026-06-26

Status: publication-candidate draft for counsel review. Do not publish until approved.

Intended URL: https://markvox.app/voice-data-policy

Summary

MarkVox uses short voice recordings only to turn spoken commands into text for the feature the user asked for.

Current voice features:

passage

arXiv ID

name folders by voice

• highlight anchor recording, where the user says the first and last words of a
• source identifier voice entry, where the user may speak a DOI, ISBN, PMID, or
• folder voice commands and folder voice titles, where the user may filter or

MarkVox does not keep voice audio after transcription. MarkVox does not sell voice data, use it for advertising, or use it for voice training.

What Is Sent For Transcription

When you start a voice feature, MarkVox records a short audio clip and sends it to OpenAI Whisper for transcription. The transcription result is returned to the app as text.

Examples:

naming

• spoken anchor words become text used for highlight matching
• a spoken source identifier becomes text used to fill the source lookup field
• a spoken folder title or command becomes text used for folder filtering or

What MarkVox Stores

MarkVox stores only the text needed for the user-facing feature.

Current storage behavior:

account export

account export

account export

title transcript may be stored so the user can filter by that folder later

• highlight anchor audio is deleted after transcription and is not stored in
• source identifier audio is deleted after transcription and is not stored in
• folder command audio is deleted after transcription and is not stored in
• folder voice title audio is deleted after transcription; the folder voice

What MarkVox Does Not Do

MarkVox does not:

• store raw voice recordings after transcription
• include voice audio files in account export
• use voice recordings or transcripts for advertising
• sell voice recordings or transcripts
• use current voice recordings for model training
• share voice data with other users

Voice Training

Voice-training recordings are not part of the current shipped product.

If MarkVox later adds a voice-training or speech-improvement feature, it must have:

• separate explicit consent
• clear retention wording
• a way to withdraw consent
• a delete path for voice-training data
• counsel review before public launch

User Controls

You can:

• decline voice recording and use typed anchors where available
• cancel recording before it is submitted
• edit or retry recognized text before saving when the app provides review
• export account data from Profile -> Data & Privacy
• request account deletion from Profile -> Data & Privacy
• contact privacy@markvox.app about voice data questions

Retention

Voice audio retention: immediate deletion after transcription attempt.

Voice transcripts:

action

deleted, or the account is deleted

• transient anchor and source-command transcripts are used for the current
• folder voice title transcripts may be stored until the folder changes, is

Account deletion should remove or anonymize stored voice-related transcripts unless limited retention is legally required.

External Service

Current transcription processor: OpenAI Whisper.

Before public launch, MarkVox must confirm current processor terms, regional/data handling options, DPA status, and any retention controls with counsel.