SayItErmano is an open-source, 100% local speech-to-text dictation application for Linux, built as a community-maintained port of the macOS app FluidVoice.

The main features of this application include:

• 100% Local Speech-to-Text: Performs all transcription locally on your machine using backends like faster-whisper (on CUDA GPU or CPU int8), whisper.cpp, PyTorch Whisper, or NVIDIA Parakeet TDT, ensuring audio never leaves your device by default.

• Universal App Insertion: Types transcribed text into any Linux application—including code editors, web browsers, terminals, and chat apps—via direct keystrokes or verified paste on both X11 and Wayland sessions.

• Spoken Punctuation & Custom Dictionary: Automatically filters out filler words ("um", "uh"), supports a custom jargon dictionary, and translates 108 spoken punctuation aliases into literal characters.

• Optional AI Polish: Provides an optional post-processing step to clean up dictation using any local or cloud OpenAI-compatible endpoint (such as Ollama, LM Studio, llama.cpp, Groq, or OpenAI).

• Strict Privacy-First Architecture: Ships with zero telemetry, opens zero network TCP listeners (communicates exclusively via a local user-owned Unix socket), and keeps API keys off disk.

• Native GTK 4 Interface & Quiet Overlay: Features native GTK 4 / libadwaita windows for settings and transcript history, accompanied by a quiet floating pill overlay that streams live text and audio waveforms while you speak.

• Voice Command Mode: Includes a dedicated hotkey mode that turns voice input into executable terminal commands with confirmation panels for safety.

• Flexible Installation: Installs without requiring root privileges via a one-shot curl script, pipx, Ubuntu .deb package, or from source.