Terms

Last updated: August 5, 2026

Speaker Scribe is free, open-source desktop software. The source code and distributed application are provided under the MIT License. The license text controls the permissions and conditions for copying, modifying, and distributing the software.

No hosted transcription service

The public website explains the project and hands the download to GitHub. It does not accept recordings, create accounts, or run transcription. The desktop application runs on your Mac and keeps its working data there.

Cloudflare hosts the website, GitHub hosts the repository and release download, Hugging Face hosts optional model files, and Apple operates macOS signature and notarization checks. Their own terms and policies apply when you use those services.

Software provided as is

As stated in the MIT License, the software is provided "as is," without warranty of any kind, including warranties of merchantability, fitness for a particular purpose, and noninfringement. The authors or copyright holders are not liable for claims, damages, or other liability arising from the software, to the extent the license provides and applicable law permits.

There is no support, uptime, compatibility, or future-release commitment.

Transcripts require review

Speech recognition can misread names, numbers, technical terms, accents, crosstalk, or poor audio. Speaker separation can merge two voices or split one voice into several labels. Cleaned text can sound confident while being wrong.

Treat output as a draft and review it against the recording before relying on it, especially in medical, legal, financial, journalistic, employment, disciplinary, accessibility, or safety-sensitive settings.

Recordings and consent are your responsibility

You are responsible for determining whether you may record, possess, process, label, disclose, or distribute the audio and resulting transcript. Recording, wiretap, privacy, employment, education, health, confidentiality, and biometric laws vary by jurisdiction and context. Some require consent from every recorded person or impose notice, retention, security, or deletion duties.

The app's local processing reduces disclosure to the project, but it does not provide consent, transfer rights from another person, or establish that a particular use is lawful. Obtain advice from a qualified professional when the legal status of a recording matters.

Speaker labels are not identity verification

Speaker Scribe distinguishes voice patterns within one recording and assigns labels such as SPEAKER_00. It does not identify an unknown person or compare a voice against a population. A real name is stored only when a user supplies it.

Whether audio or a derived representation is legally treated as biometric or other protected information depends on the governing law, the technology, the purpose, and the surrounding facts. Do not treat the app's product terminology as a legal classification.

Third-party components and models

The application bundles open-source components under their respective licenses, including mlx-whisper, Silero VAD, SpeechBrain, PyTorch, CPython, and ffmpeg. Model weights and optional downloads may have separate licenses or terms set by their publishers. Anyone redistributing a modified bundle is responsible for reviewing the licenses and notices that apply to that bundle.

The packaged ffmpeg build is configured without GPL components. A distributor who changes that build must reassess the resulting licensing obligations.

Changes

These terms and the privacy notice are versioned with the source. A future release may update them prospectively. The license accompanying a particular copy of the software continues to govern that copy.