Privacy

Last updated: August 5, 2026

Speaker Scribe is a local-first desktop application. The app processes recordings on your Mac; it is not a hosted transcription service. The project website does not have accounts, forms, advertising, or application analytics. Internet delivery still involves the providers named below, so this notice does not make the inaccurate claim that visiting the website creates no network data.

Scope and contact

This notice covers the Speaker Scribe desktop application and the public project website. The open-source project is maintained through the rogu3bear/speaker-scribe repository.

For a private privacy or security report, use GitHub's private vulnerability reporting form. Do not put recordings, transcripts, access tokens, or other sensitive personal information in a public issue.

Desktop application data

When you import a recording, the application keeps the audio, transcript, speaker labels, job status, and exports on your Mac. The packaged application uses:

~/Library/Application Support/com.mlnavigator.speaker-scribe/

A source checkout uses its local data/ directory unless configured otherwise. Deleting the applicable data directory removes the jobs and copied recordings kept there. Exported files remain wherever you saved them.

Speaker diarization temporarily computes numerical embeddings from short speech windows so it can group similar voices within one recording. Those working embeddings are used in memory and are not written to the job record. The saved record contains speaker labels and transcript segments, not a reusable voice profile. A label such as SPEAKER_00 means a distinct voice within that job; it does not identify a person. Real names appear only when you enter them.

Depending on the jurisdiction and how audio or derived features are used, recordings or voice-derived data may be regulated as personal or biometric information. Speaker Scribe's local processing does not decide whether you have the rights or consent required for a particular recording.

Desktop network activity

Transcription and diarization do not upload your audio or transcript. The packaged application includes the models needed for its default workflow.

If you choose Download for another speech model, the app requests that model from Hugging Face. Hugging Face receives ordinary connection and request metadata, which can include an IP address, device or software information, time, and the requested model. Audio, transcripts, and speaker labels are not part of that model request. Downloaded model files remain in the local model cache until you remove them.

macOS may also verify the application's Developer ID signature and notarization through Apple's Gatekeeper mechanisms. Those operating-system checks do not send your recordings to Speaker Scribe.

Website data and providers

The website code does not set a session or tracking cookie, use browser storage, run advertising or fingerprinting code, accept uploads, or send application analytics. It has no application database. The following infrastructure and external links nevertheless process ordinary web requests:

• Cloudflare: contacted when the website or an asset is requested. It processes IP addresses, routing and request metadata, system configuration information, and security/network data needed to deliver and protect the site.
• Cloudflare Network Error Logging: if the browser supports reporting and a sampled network failure occurs, it can send a browser-generated failure report. Cloudflare says it derives coarse network/location data and discards the client IP after processing that report.
• GitHub: contacted when you follow the source link or start the direct DMG download. GitHub processes IP address, device and request information, and other service-usage data under its policy.
• Hugging Face: contacted only when you choose an optional model download in the app. Hugging Face processes the IP address, requested model, device/software and request metadata under its policy.
• Apple: macOS can perform its own signature or notarization checks. Apple handles those checks under its settings and policies; Speaker Scribe does not supply Apple with a recording or transcript.

The project does not sell personal information, serve behaviorally targeted ads, or use these requests to build visitor profiles. Each provider controls its own retention, legal bases, and privacy-request process as described in its linked policy.

Retention and deletion

• Jobs and copied recordings remain in the local application data directory until you remove them.
• Files you export remain at the destination you selected.
• Optional model files remain in their local caches until you remove them.
• Temporary diarization embeddings are discarded after the job and are not stored as reusable voice profiles.
• The project website has no application database or account record to delete. Cloudflare, GitHub, Hugging Face, and Apple apply their own retention rules to the network data they process.

Because the project does not maintain website visitor accounts or a project-controlled visitor database, the maintainer normally cannot identify a website visitor in order to retrieve or delete a visitor profile. Questions about provider-held data should be directed to the relevant provider. Questions about data found in the local application can be sent privately through the contact route above.

Children

Speaker Scribe is a general-purpose transcription tool, is not directed to children, and does not provide accounts or a project-operated service that asks children for personal information.

Changes

This notice is versioned with the application. Its repository history records changes. A material change to collection or network behavior requires a matching code, documentation, and notice review before release.

Legal scope

This notice describes the project's verified behavior; it is not a guarantee that one notice satisfies every obligation for every user, recording, or jurisdiction. People using recordings remain responsible for the laws and consents applicable to their own activity.