Murmur v1.3.0 — Studio

Fully local Windows dictation — press F1, talk, it types. No Python, no dependencies, no cloud.

Install

1. Download Murmur-release.zip below and extract it to a user-writable folder (not Program Files)
2. Run Murmur.exe — first run downloads a Whisper model into models/
3. Press F1, talk, pause — it types into whatever window is focused

Already have Murmur v1.2.4+? You don't need to download anything — Murmur updates itself. Click the "Update to v1.3.0" button in the menu bar (or Help → Check for Updates).
SmartScreen: the exe is not code-signed yet — click More info → Run anyway.

New in 1.3.0 — the Studio look

• A designed visual identity: unified teal accent across the whole app, a living hero banner (breathing outline when idle, an audio-reactive glow that follows your voice while recording), a glowing gradient spectrum analyzer, smooth-gradient level meter, pill-shaped mode buttons, status chip with live phase dot, and accent-marked sections.
• Prefer the old look? View → Theme → Classic brings back the original style; your choice is remembered.
• Latency readout is now a clean right-aligned table.

Recent

• 1.2.4: one-click self-updates, dark title bar
• 1.2.3: DSP slider fixes, simpler audio section
• 1.2.1–1.2.2: model picker + hardware auto-detect, full feature-audit hardening

MIT licensed. Report issues at https://github.com/Roach9223/Murmur/issues

Murmur v1.2.4

Fully local Windows dictation — press F1, talk, it types. No Python, no dependencies, no cloud.

Install

1. Download Murmur-release.zip below and extract it to a user-writable folder (not Program Files)
2. Run Murmur.exe — first run downloads a Whisper model into models/
3. Press F1, talk, pause — it types into whatever window is focused

SmartScreen: the exe is not code-signed yet — click More info → Run anyway.
Murmur-update.zip is for Murmur's built-in updater — new users want Murmur-release.zip.

New in 1.2.4 — one-click updates

• Murmur now updates itself: when a new version ships, an "Update to vX.Y.Z" button appears in the menu bar. One click downloads it (usually ~25MB, not the full package), and "Restart & Update" installs and relaunches automatically. Settings, models, and recordings are always kept. Manual check: Help → Check for Updates.
• Dark title bar — the window chrome finally matches the dark UI.

Recent

• 1.2.3: fixed DSP sliders snapping back; simpler audio section; visible slider lock
• 1.2.2: full feature audit — no more lost transcriptions, honest calibration, MP3 export
• 1.2.1: Whisper model picker + hardware auto-detect for AMD/CPU machines
• 1.2.0: Raw mode default, redesigned UI, self-contained ~1GB download

MIT licensed. Report issues at https://github.com/Roach9223/Murmur/issues

Murmur v1.2.3

Fully local Windows dictation — press F1, talk, it types. No Python, no dependencies, no cloud.

Install

1. Download Murmur-release.zip below and extract it to a user-writable folder (not Program Files)
2. Run Murmur.exe — first run downloads a Whisper model into models/
3. Press F1, talk, pause — it types into whatever window is focused

SmartScreen: the exe is not code-signed yet — click More info → Run anyway.

New in 1.2.3

• Fixed: DSP sliders snapped back to their previous value on release (regression in 1.2.2) — threshold and floor adjustments now stick
• Simpler audio section: the compressor UI is retired (Whisper doesn't need compression; it remains available via config.json for power users, with an indicator when active)
• The slider lock is now a clearly visible colored button directly above the sliders (amber = locked, green = unlocked)

Recent

• 1.2.2: full feature audit — no more lost transcriptions, hideable transcription progress, honest calibration, MP3 export + record source in the UI
• 1.2.1: Whisper model picker + hardware auto-detect (AMD/CPU machines get a fast model automatically)
• 1.2.0: Raw mode default (no LLM required), redesigned UI, self-contained ~1GB download

MIT licensed. Report issues at https://github.com/Roach9223/Murmur/issues

Murmur v1.2.2

Fully local Windows dictation — press F1, talk, it types. No Python, no dependencies, no cloud.

Install

1. Download Murmur-release.zip below and extract it to a user-writable folder (not Program Files)
2. Run Murmur.exe — first run downloads a Whisper model into models/
3. Press F1, talk, pause — it types into whatever window is focused

SmartScreen: the exe is not code-signed yet — click More info → Run anyway.

New in 1.2.2 — hardening from a full feature audit

• Transcriptions can no longer be lost: a failed save (bad filename, disk issue) keeps the transcription and shows why; filenames are sanitized; existing transcripts are never silently overwritten
• File transcription no longer locks the app: hide the progress popup and keep working; a live progress button brings it back
• Mic calibration is honest: refuses to "succeed" if you didn't speak, stops dictation first (no more typing the calibration sentence into your editor), and can be cancelled
• New controls: Pre/Post-DSP record source selector, one-click MP3 export for recordings, tooltips on every audio control
• Smoother: DSP sliders no longer spam the engine while dragging, meters stay live with the gate off, recording failures (disk full) are reported instead of freezing the timer, Diagnostics shown by default

Recent

• 1.2.1: Whisper model picker + hardware auto-detect (AMD/CPU machines get a fast model automatically)
• 1.2.0: Raw mode default (no LLM required), redesigned UI, self-contained ~1GB download

MIT licensed. Report issues at https://github.com/Roach9223/Murmur/issues

Murmur v1.2.1

Fully local Windows dictation — press F1, talk, it types. No Python, no dependencies, no cloud.

Install

1. Download Murmur-release.zip below and extract it to a user-writable folder (not Program Files)
2. Run Murmur.exe — first run downloads a Whisper model into models/
3. Press F1, talk, pause — it types into whatever window is focused

SmartScreen: the exe is not code-signed yet — click More info → Run anyway.

New in 1.2.1 — better for AMD / CPU-only machines

• Whisper model picker in the UI: switch models from a dropdown (Distil Large v3.5, Large v3, Turbo, Medium/Small/Base EN) — no config editing. Models download automatically on first use.
• Hardware auto-detect: NVIDIA GPU → the big distil model on CUDA. AMD/Intel/no GPU → small.en on CPU automatically, which is near-real-time on a modern CPU (previously the large model crawled at several seconds per phrase).
• GPU/CPU badge shows which hardware is actually transcribing. AMD/Intel GPUs can't accelerate inference yet (CUDA-only engine); a Vulkan backend is planned.

From 1.2.0

• Raw mode default (no local LLM required), redesigned UI, CPU fallback, orphan-process and crash fixes, self-contained ~1GB download.

MIT licensed. Report issues at https://github.com/Roach9223/Murmur/issues

Murmur v1.2.0

Fully local Windows dictation — press F1, talk, it types. No Python, no dependencies, no cloud.

Install

1. Download Murmur-release.zip below and extract it to a user-writable folder (not Program Files)
2. Run Murmur.exe — first run downloads the Whisper model (~1.5GB) into models/
3. Press F1, talk, pause — it types into whatever window is focused

SmartScreen: the exe is not code-signed yet — click More info → Run anyway.
No NVIDIA GPU? Transcription falls back to CPU automatically (slower, but works).

What's new

• Works out of the box: Raw mode is now the default — no LM Studio or local LLM required. Clean/Prompt/Dev/Detailed modes light up if you run one.
• Redesigned UI: one-click mode pills, profile + mic pickers on the main surface, "Heard/Typed" output front and center, collapsible sections for DSP/recording/diagnostics. The spectrum analyzer and VU meter stay full width.
• Reliability: CPU fallback without an NVIDIA GPU, no more orphaned engine processes, engine-crash errors surfaced in the UI, no UI freezes on slow operations, non-ASCII install paths supported, and a long list of engine fixes from a full pre-release audit.
• Self-contained ~1GB download: Python runtime, CUDA/cuDNN, and the C++ runtime are all bundled.

MIT licensed. Report issues at https://github.com/Roach9223/Murmur/issues

Murmur v1.1.0

This release was published without a downloadable build. Use v1.2.0 instead, which includes the self-contained Windows download.