joewongjc/type4me

Type4Me – macOS Voice‑First Input Method

What it is – A macOS‑only application that lets you dictate text, translate, run prompts, and even control the computer by voice. It combines local speech‑recognition engines with optional cloud ASR services and can forward the transcribed text through LLMs for polishing, translation, or task generation.

Core capabilities

• Speech‑to‑text: built‑in local recognizer (SenseVoice + Qwen3‑ASR) plus support for 15 cloud ASR providers; streaming output while you speak.
• Eight ready‑to‑use modes: Quick (raw dictation), IntelliSense (context‑aware polishing), Translation, Ask‑Anything, macOS Actions, Voice Polish, Prompt Optimizer, Task‑Delegation. Each mode can be bound to multiple global hotkeys (hold‑to‑talk or toggle).
• LLM integration: Connect to major cloud LLM APIs or a local Ollama server; unified UI for credentials, testing, and default engine selection.
• Vocabulary management: Hot‑words and snippet replacement to fix proper‑noun errors; a companion "Vocabulary Management Skill" can auto‑manage these rules.
• Menu‑bar control center: Switch modes, microphones, ASR/translation targets, view status, history, and updates.
• History & export: All raw and processed transcriptions are saved, can be tagged, inspected, and exported as CSV.
• URL scheme: type4me:// commands let external tools (Raycast, Alfred, Stream Deck, scripts) start/stop recordings, open settings, or edit vocabularies without stealing focus.
• Voice‑revise: After a dictation, press a shortcut and speak a correction; the app replaces the targeted slot and offers one‑click undo.

Typical workflow

1. Press a hotkey (e.g., Fn) → start recording.
2. Speak; the engine streams text to the active cursor.
3. Depending on the selected mode, the raw text may be sent to an LLM for polishing, translation, or task generation before being injected.
4. Use the Ask‑Anything window to query selected text or continue a conversation; history is searchable.
5. Control macOS (open apps, adjust volume, take screenshots, etc.) entirely by voice.

Who it’s for

• Professionals who spend a lot of time typing on macOS and want hands‑free input.
• Developers and writers needing fast dictation, automatic polishing, or on‑the‑fly translation.
• Users who want a customizable voice interface that can also trigger macOS actions.

Installation

• Two DMG builds (cloud‑enabled ≈ 10 MB, local‑engine ≈ 700 MB). Both share the same configuration folder, so you can switch at any time.
• Requires macOS 14+ (Sonoma) on Intel or Apple Silicon. The local build needs ~8 GB RAM for SenseVoice and ~8 GB for Qwen3‑ASR.
• After download, open the DMG, drag Type4Me to Applications, then grant microphone, accessibility, and automation permissions.

Extensibility

• ASR providers follow a plug‑in interface (ASRProviderConfig + SpeechRecognizer). Adding a new service is a matter of registering the provider.
• Custom modes are created by defining a prompt template that can reference {text}, {selected} and {clipboard} placeholders.

License – MIT License.

---

All information above is taken directly from the repository’s README; no additional features have been inferred.