Transcribe Offline

Transcribe Offline is an open-source desktop app for local transcription, speaker diarization, and transcript review/edit workflows. It is implemented as a native Rust/egui GUI on top of Openresearchtools-Engine.

• Windows x64*: Transcribe-Offline.exe
• macOS arm64*: Transcribe-offline.dmg
• Linux amd64: install transcribe-offline through APT.

Install on Debian 13+ or Ubuntu 24.04/26.04+

For the first Open Research Tools installation on a system, copy and run this one command:

wget -qO /tmp/keyring.deb https://keyring.openresearchtools.com && sudo apt install -y /tmp/keyring.deb && sudo apt update && sudo apt install -y transcribe-offline

If the Open Research Tools APT repository is already configured:

sudo apt install transcribe-offline

Linux engine packages

The Linux app package uses system-installed Openresearchtools-Engine runtimes and depends on both APT packages:

• openresearchtools-engine for Vulkan (/opt/openresearchtools/engine/vulkan)
• openresearchtools-engine-cuda for CUDA (/opt/openresearchtools/engine/cuda)

The Vulkan/CUDA selector switches directly between those locations. Linux does not download, unpack, unblock, or replace the engine from inside the app. The APT installation above resolves both engine dependencies automatically.

NOW AVAILABLE! Realtime Live Transcription + Live Speaker Diarization

Transcribe Offline now supports true realtime live transcription with true realtime live speaker diarization.

This is not a batch-style “record a chunk, stop, and transcribe later” workflow. The live pipeline is designed around continuous streaming audio, continuous session state, and sliding-context realtime model execution so transcription and speaker diarization can update as speech is happening.

This was made possible by the native runtime work in Openresearchtools-Engine, which extends a ggml / llama.cpp-based backend with the in-process bridge, streaming audio session handling, realtime decoding logic, and backend support needed to run these models locally with CUDA, Vulkan, and Metal acceleration.

For live transcription, this app uses converted realtime GGUF artifacts published at openresearchtools/Voxtral-Mini-4B-Realtime-2602, based on the upstream Voxtral realtime model from mistralai/Voxtral-Mini-4B-Realtime-2602.

For live speaker diarization, this app uses converted GGUF artifacts published at openresearchtools/diar_streaming_sortformer_4spk-v2.1-gguf, based on the upstream NVIDIA Sortformer model from nvidia/diar_streaming_sortformer_4spk-v2.1.

We are grateful to the original model creators and upstream projects for making these capabilities possible. Our use, conversion, packaging, and local runtime integration of these models does not imply endorsement, sponsorship, or affiliation by upstream authors or maintainers of packages and models.

What it does

• transcription workflows (speech / subtitle / transcript),
• realtime live transcription with optional live speaker diarization,
• speaker diarization,
• local transcript chat and anonymisation,
• and desktop-first editing/playback flows.

1) Local transcription modes

Runs the engine audio path for:

• speech output,
• subtitle output,
• transcript output with speaker diarization.

2) Speaker diarization

Uses the engine's native realtime Sortformer diarization path together with the offline transcript assembly/sanitization flow.

3) Transcript review and editing

Provides side-by-side transcript/edit views, playback-linked navigation, autosave, speaker rename tools, and anonymisation pass tooling.

4) Local LLM helpers

Uses the engine bridge chat path with local GGUF models for transcript Q&A and anonymisation extraction. For chat/anonymisation, you can use any llama.cpp-compatible GGUF model; instruction-following chat models are recommended.

Anonymisation is a beta function. For real-world use, always manually review the output transcript to confirm no unintended sensitive data remains.

Designed to be local‑first. However, no software can guarantee absolute privacy or security. Please consider your threat model and institutional policies before processing sensitive material.

Unsigned Build Notice

This app is an open-source hobby development effort by the repository owner. We do not currently have funding for full paid code-signing and notarization pipelines across all platforms/releases.

Because of that, operating-system protections or hardened security environments (for example Windows SmartScreen, enterprise endpoint controls, or macOS Gatekeeper policies) may block unsigned binaries.

If your environment blocks unsigned binaries, the recommended path is:

• build this desktop app from source on the target device,
• build Openresearchtools-Engine from source on the same target device,
• and use those locally-built artifacts in your deployment.

Windows (when blocked)

• If SmartScreen shows "Windows protected your PC", use More info -> Run anyway only if your policy allows it.
• In the app, go to Settings -> Runtime Setup and run:
  • Download/Repair runtime
  • Unblock unsigned runtime
  • Recheck
• The Windows unblock script clears Mark-of-the-Web flags in the selected runtime directory by running Unblock-File recursively on runtime files.

macOS (when blocked)

• Try Right click -> Open on first launch.
• If blocked by Gatekeeper, use System Settings -> Privacy & Security -> Open Anyway when available and policy permits.
• In the app, after runtime install/repair, click Unblock unsigned runtime then Recheck.
• The macOS unblock script removes quarantine attributes recursively (xattr -dr com.apple.quarantine) and restores executable bits for runtime binaries/scripts where needed (chmod +x on relevant files).

---

Highlights

• Offline-first runtime flow with in-app runtime install/repair on Windows/macOS and APT-managed Vulkan/CUDA runtimes on Linux.
• Native desktop orchestration of Openresearchtools-Engine (llama-server-bridge).
• Single device selection model (CPU or selected GPU) for runtime execution.
• Built-in transcript editing, playback follow, anonymisation, and export workflow.

---

How it works (in this repo)

• This app is a GUI/orchestration layer.
• Openresearchtools-Engine provides the local runtime components.
• The app invokes runtime features through llama-server-bridge.
• Playback decode in the app uses the Rust Symphonia stack; runtime-side media conversion uses engine FFmpeg components.

Project relationship

• Transcribe Offline is a reference example of integrating Openresearchtools-Engine in a native desktop GUI.
• This app uses llama-server-bridge from Openresearchtools-Engine.
• Openresearchtools-Engine runs on a modified llama.cpp runtime path with native Whisper transcription, Voxtral realtime transcription, and Sortformer diarization integrations.
• This desktop app itself is a wrapper/orchestrator around that runtime.
• This desktop app relies on Openresearchtools-Engine runtime media components (FFmpeg/PDFium) at runtime.

Acknowledgements (used in this app/runtime path)

• Openresearchtools-Engine: embeddable runtime used by this app (llama-server-bridge, runtime orchestration, and model/device execution path).
• egui / eframe: native immediate-mode GUI framework used to build this desktop application UI.
• llama.cpp and ggml: core inference runtime and device/offload mechanics used through Openresearchtools-Engine.
• whisper.cpp: transcription backbone used by the engine audio pipeline.
• mistralai/Voxtral-Mini-4B-Realtime-2602: upstream realtime speech-to-text model reference used for the app's live transcription path.
• nvidia/diar_streaming_sortformer_4spk-v2.1: upstream realtime diarization model reference used for the app's live diarization path.
• Qwen/Qwen3.5-9B: upstream local chat/anonymisation model family reference used by the app's managed GGUF downloads.
• FFmpeg (LGPL shared runtime builds): media decoding/normalization path used by Openresearchtools-Engine runtime for audio conversion.
• Symphonia: Rust audio decoding stack used by this desktop app for local playback.

Model acknowledgements (upstream model repos)

• openresearchtools/whisper-large-v3-turbo-GGML and openresearchtools/whisper-large-v3-GGML: converted Whisper runtime artifacts used by the app's managed transcription downloads.
• openai/whisper-large-v3-turbo and openai/whisper-large-v3: upstream Whisper model family references for the managed transcription downloads.
• openresearchtools/Voxtral-Mini-4B-Realtime-2602: converted realtime Voxtral artifacts used by the app's live transcription downloads.
• mistralai/Voxtral-Mini-4B-Realtime-2602: upstream Voxtral realtime model family reference for the managed live downloads.
• openresearchtools/diar_streaming_sortformer_4spk-v2.1-gguf: converted Sortformer diarization artifact used by the app's live diarization path.
• nvidia/diar_streaming_sortformer_4spk-v2.1: upstream Sortformer diarization model reference for the managed live downloads.
• openresearchtools/Qwen3.5-9B-GGUF: converted GGUF chat/anonymisation artifacts used by the app's local chat helpers.
• Qwen/Qwen3.5-9B: upstream chat model family reference for the managed local LLM downloads.

Non-endorsement statement

This project is independent work by OpenResearchTools. This project is not affiliated with, sponsored by, or endorsed by the maintainers/owners of any third-party projects listed above or in the bundled notices/license files.

All third-party names and marks remain property of their respective owners.

Licensing and notices

The Transcribe Offline application source code is licensed under the MIT License; third-party dependencies and bundled components remain licensed under their respective original licenses.

Read these files in this repo:

• Notice page (app + models + engine): licenses/THIRD_PARTY_NOTICES_ALL.md
• Full app third-party licenses (full text per package): licenses/THIRD_PARTY_LICENSES_ALL.md
• Full engine third-party licenses (full text per package/file): licenses/ENGINE_THIRD_PARTY_LICENSES_FULL.md
• Help -> Notices
• Help -> App licenses
• Help -> Engine licenses

Models and conversions

• Whisper model binaries are fetched from openresearchtools/whisper-large-v3-turbo-GGML and openresearchtools/whisper-large-v3-GGML.
• Live transcription model binaries are fetched from openresearchtools/Voxtral-Mini-4B-Realtime-2602.
• Live diarization model binaries are fetched from openresearchtools/diar_streaming_sortformer_4spk-v2.1-gguf.
• Local chat/anonymisation GGUF models are fetched from openresearchtools/Qwen3.5-9B-GGUF.
• OpenResearchTools publishes converted model artifacts for runtime compatibility.

Converted-model note:

• These converted artifacts are provided for interoperability with this runtime.
• They are not upstream-official releases and are not endorsed by upstream model owners.
• Use follows the original model licenses and model-card terms.

For citations and model lineage references, see the bundled Notices document:

• licenses/THIRD_PARTY_NOTICES_ALL.md

How to cite

Suggested citation:

Rutkauskas, L. (2026). Transcribe Offline (Version 2.1.0) [Computer software]. OpenResearchTools. https://github.com/openresearchtools/transcribeoffline.

BibTeX:

@software{Rutkauskas_TranscribeOffline_2026,
  author    = {Rutkauskas, L.},
  title     = {Transcribe Offline},
  version   = {2.1.0},
  date      = {2026-03-04},
  url       = {https://github.com/openresearchtools/transcribeoffline},
  publisher = {OpenResearchTools},
  license   = {MIT}
}