TypeWhisper: Private Local Speech-to-Text for Mac & Windows

TypeWhisper (official site) is a local-first speech-to-text app that runs speech recognition on your own desktop. The macOS version is at STABLE 1.5, the Windows version is STABLE, and an iOS version is in ALPHA — it's made in Germany. Press a global hotkey, speak, and the finalized text is inserted directly into whatever app has focus, functioning as a system-wide dictation tool.

The core workflow is simple: press a configured global hotkey, speak, and the recognized text is inserted at the cursor position in whichever app is active — email client, chat, a document editor, a browser form — without needing to open TypeWhisper itself first.

- System-wide dictation: press a global hotkey and speak; text is inserted directly at the cursor in any app
- File transcription: turn audio files into text
- Workflows: reorder processing steps via drag and drop, or set up watched-folder jobs
- Dictionary, snippets, and history: register frequent phrases and review past conversions
- Per-dictation engine overrides: switch recognition engine/model per dictation
- Local HTTP API: automate with external tools

Beyond live dictation and file transcription, TypeWhisper also supports multiple concurrent global hotkeys and automation via its local HTTP API.

Version 1.5 added app-aware dictation, dictionary learning, local model memory, cloud ASR upload, and a fullscreen indicator, among other enhancements.

TypeWhisper is built to be local-first: speech recognition runs locally on the desktop, with no telemetry, no subscription, and no required cloud dependency. Its core is open source (1.8k stars on GitHub), and the core app itself is free to use.

The only paid offering is a Commercial license — a one-time purchase, not a subscription. It unlocks two features: Cloud Folder Sync, which syncs the dictionary and snippets across devices using your own iCloud Drive, Dropbox, OneDrive, or Syncthing storage (no TypeWhisper account required), and Automatic Correction Learning, which learns from high-confidence corrections made after text is inserted — recurring names, jargon, and typos — and applies them automatically going forward. The core app remains free, open source, and local-first without either of these.

TypeWhisper offers a plugin hub combining official plugins, community plugins, and manually added ones, with an SDK available for building your own. Notable examples include Groq, which provides very fast cloud transcription and LLM processing via its LPU, and NVIDIA Parakeet TDT, a bundled transcription engine optimized for local use on Apple Silicon. Users can switch between cloud speed and fully local processing on a per-plugin basis depending on the task.

Speech-to-text options broadly fall into three categories: built-in OS dictation, cloud-based speech recognition services, and local-first apps like TypeWhisper. Here's how they compare on a few practical axes.

Stated plainly: TypeWhisper defaults to local processing while still offering cloud-based plugins like Groq as an option, and it uses a one-time Commercial license rather than a subscription. Which option fits best depends on how sensitive the content is, how much speed matters, and how well it fits an existing workflow.

Dictating meeting notes, internal memos with client or project names, or anything touching contract terms are common cases where you'd rather not send content off-device. Cloud speech recognition services typically send audio to an external server, and how that data is handled depends on that service's terms and infrastructure. Because TypeWhisper defaults to local speech recognition, it offers an option for not sending that audio to an external server — though that changes if a cloud-based plugin like Groq is enabled, since the engine choice is made per plugin.

For a related tool focused on local transcription on Apple Silicon, see Qwen Scribe (local dictation). Which tool fits better depends on whether the priority is transcription specifically or system-wide dictation.

TypeWhisper is a local-first speech-to-text app: press a global hotkey, speak, and text is inserted directly at the cursor. Its core is free and open source, with no telemetry and no subscription, while a one-time Commercial license unlocks just two extras — Cloud Folder Sync and Automatic Correction Learning. Its plugin hub adds cloud engines like Groq as an option, so the choice between local processing and cloud speed can be made per task.