murmur v0.1.1

Quality and accuracy improvements over v0.1.0. Fully local, offline, and free by default.

Fixed

• First recording after boot is no longer truncated. The microphone is now warmed at startup, so a cold audio device no longer drops the start of your first dictation.
• No more dropped spaces. Typed text is now paced so the target window can't drop keystrokes (most often spaces). Tunable via inject_char_delay_ms (0 = instant/unpaced).
• GPU (DirectML) provider reliability — always launches the correct sidecar.

Added

• Gaming / internet slang recognition — LARP, NPC, AFK, MMO, MOBA, RPG, DPS, FPS, RNG, IMO, and more get correct casing, with guards so ordinary words ("ring", "lark") are never mangled.
• Slovenian + auto-detect language support ("language": "sl" or "auto"); cleanup no longer forces English.
• Built-in brand / dev-term fixes (GitHub, GitLab, Roblox, PyInstaller, Luau, Rojo, …) applied on every path.

Notes

• Optional cloud speech-to-text (Groq / OpenAI) and AI cleanup (Cerebras / Groq / Anthropic) are available with a free API key — your audio stays local unless you opt in.
• Windows 10 / 11, x64. Download the .msi below. WebView2 ships with current Windows; the speech model downloads automatically on first use, then runs offline.