Why Murmur?

Most transcription tools upload your recordings to external servers. Meetings, interviews, confidential conversations — everything ends up in a cloud you don't control. Processing happens on hardware that belongs to someone else. Your data is stored indefinitely, sometimes used to train AI models.

For many use cases, that's a problem. Confidential business discussions, research interviews with sensitive data, medical dictations, legal consultations — anywhere privacy isn't optional.

Murmur takes a different approach.

Murmur downloads two AI models on first launch — a speech recognition model (Parakeet ASR, ~150 MB) and a language model for summaries (Qwen 3.5, ~4 GB). After that, the app doesn't need an internet connection for its core work.

Speech recognition runs on your Apple Silicon chip's Neural Engine and GPU. Summaries are generated by a local language model with 9 billion parameters — directly on your Mac. No API calls, no cloud processing.

Recordings and transcripts are encrypted with AES-GCM before they're written to disk. There are no user accounts, no telemetry, no data collection. The only network requests: license checks, software updates, and the one-time model download.

€29 one-time for version 1 — no subscription, no hidden fees. The license covers up to three Macs. Two months of a cloud transcription service (€15-20/month) costs the same as a permanent Murmur license.

The 7-day trial includes full functionality. No features held back. After the trial, you can still open and export your existing transcriptions. 30-day money-back guarantee after purchase.

Murmur is a macOS app for Apple Silicon. No Windows, no Linux, no Intel Macs. If you have a team with mixed hardware, Murmur only works for the Mac users.

There's no real-time collaboration and no browser-based interface. If your workflow depends on multiple people editing a transcript simultaneously, cloud services are the better choice.

Murmur supports 25+ languages — but if you work with rare languages or dialects, specialized cloud services with larger models may be more accurate.