Desktop voice dictation for Windows and macOS. Speak, transcribe, and insert text into the app you're working in—with local speech recognition or your own cloud API keys.
Download 0.1.1 · Quick start · Screenshots · Contribute
No Murmur account, activation server, or subscription is required. Local recognition is the default; optional cloud services use your credentials and may charge their own fees.
0.1.1 is an early preview. Choose the build for your computer:
| Platform | Download |
|---|---|
| Windows x64 | Windows installer (.exe) |
| macOS · Apple Silicon | Mac application (.zip) |
| macOS · Intel | Mac application (.zip) |
Release notes · SHA-256 checksums
These binaries are unsigned and not notarized. Windows SmartScreen or macOS Gatekeeper may block them. Do not disable system-wide security protections. If you prefer not to approve an unsigned application, build from the reviewed source instead. Users of 0.1.0 should upgrade before using the experimental ChatGPT integration.
- Dictate across apps. Start and stop with a configurable global shortcut; insert recognized text through the clipboard.
- Local speech recognition. Download a Whisper model and its runtime, then use local recognition without cloud transcription.
- Your choice of providers. Optional Groq, Deepgram, AssemblyAI, and ElevenLabs speech adapters use your own API keys.
- Optional AI rewriting. Apply explicit AI modes to dictated text with your own provider key.
- Local history and statistics. Review previous dictations and usage on your computer.
- Desktop controls. Configure microphone, shortcuts, sounds, and interface language.
No marketing telemetry, commercial quotas, or automatic update polling. The community application has its own identity and data directory; it does not import or overwrite the previous commercial application's data.
Actual application captures from the isolated Windows Electron smoke test, using a clean English-language profile. No personal transcripts or provider credentials are present. These show the interface, not a live microphone or cloud-provider demonstration. Capture details
Choose local recognition or a cloud provider, and open the local model manager. Key fields are empty.
Optional post-processing, custom modes, dictionary, and snippets. Post-processing is off in this clean profile.
- Install the Windows build, or extract the Mac ZIP and move the application to Applications.
- Complete setup. In Transcription → Manage Models, install a local model and runtime. These are separate, on-demand downloads—not bundled with the app. Check the model's language support and allow sufficient disk space. Model requirements
- Select your microphone and dictation shortcut. Grant microphone access when requested; macOS may also require Accessibility permission for text insertion.
- Focus a text field in another app. Use your configured shortcut to start dictation, speak, then use it again to stop and insert the transcript.
For local-only use, keep AI post-processing off as well as selecting local speech recognition. Cloud transcription sends audio to the selected provider; cloud AI processing sends text. History is stored locally and is not an encrypted transcript vault. Privacy and network behavior
Requires Windows x64 or macOS, Node.js 22.12+ in the 22.x line (recommended), and npm. No .env, backend, signing identity, or Murmur account is needed.
git clone https://github.com/airope/murmur-community.git
cd murmur-community
npm ci
npm run devThe default branch may differ from the downloadable preview. To reproduce 0.1.1, check out v0.1.1 before running npm ci.
Built with Electron, React, and TypeScript, with shared Windows/macOS code and separate provider adapters. Architecture and trade-offs
npm run typecheck
npm test
npm run build
npm run test:smoke
# Package on the corresponding operating system
npm run package # Windows
npm run package:mac # macOSCI checks Windows/macOS builds, isolated Electron startup, settings IPC, and real local recognition of a public audio fixture. The smoke test substitutes native shortcut/login-item side effects. It does not certify physical microphone capture, paste into every application, or authenticated cloud-provider behavior. Verification scope
- Early preview with best-effort maintenance, not a support SLA. Linux is not a supported release target.
- Local models need disk space and CPU resources; language coverage and recognition quality vary. No speed or accuracy benchmark is claimed.
- The ChatGPT website adapter is experimental, not an official API or an OpenAI-endorsed integration. Website changes and restricted login flows can break it; prefer local recognition or supported APIs.
- Automatic active-window context detection is deliberately excluded. Explicit AI modes remain available.
- Packaging for both Mac architectures does not establish runtime testing on both. Hardware, permissions, keyboard layouts, and live provider compatibility need testing on your system.
Bug reports, focused fixes, documentation, and Windows/macOS compatibility reports are welcome. Include your OS, app version, steps to reproduce, and sanitized logs—never API keys or personal transcripts. Add a regression test for bug fixes and run the checks above.
Read CONTRIBUTING.md before opening a pull request. Report sensitive issues through SECURITY.md, not a public issue.
Murmur began as a personal desktop product. This community edition separates the desktop application from the private website, billing, and account services. Development has been AI-assisted.
The current source is available under the MIT license. The already-published 0.1.1 release was distributed under MIT; its tagged license remains the reference for that release. Third-party libraries, model weights, runtimes, and APIs retain their own licenses and terms. Third-party notices