VoiceTypr

VoiceTypr is an open-source desktop dictation tool that converts speech into text inside any running application. It runs Whisper speech recognition models locally on user hardware instead of streaming audio to external servers. A system-wide shortcut inserts the transcription directly at the active cursor position in code editors, terminals, and chat applications.

Why VoiceTypr instead of Wispr Flow

Wispr Flow routes microphone audio to cloud servers and bills users on a monthly subscription. For engineers dictating private code, internal API tokens, system hostnames, or production incidents, sending live voice streams outside company infrastructure introduces serious compliance risks. VoiceTypr handles speech recognition offline using local Whisper models accelerated by Apple Silicon Metal or Windows GPUs. Audio never leaves the workstation, keeping sensitive technical context confined to the local machine.

Control also extends to transcription cleanup. While Wispr Flow formats text through proprietary hosted pipelines, VoiceTypr treats generative enhancement as an optional, bring-your-own-key feature. Developers can route transcribed text—never raw audio—through Groq, Anthropic, or OpenAI to produce structured git commits and markdown notes. Teams operating under zero-exfiltration policies can keep cloud integrations disabled and rely solely on offline transcription.

What it does well

• System-wide hotkey dictation inserts text directly at the cursor in tools like Cursor, VS Code, and terminal emulators without window switching.
• Local model execution uses Apple Silicon Metal or Windows GPU acceleration to deliver responsive transcription with zero network latency.
• Tiered model selections let developers balance disk footprint against accuracy, supporting compact Whisper checkpoints up to larger models.
• Optional AI formatting modes convert unedited spoken dictation into structured git commit messages, pull requests, or notes via personal API keys.
• The built-in CLI allows developers to script batch audio transcription directly inside terminal workflows and local automation pipelines.
• Cross-device network pairing allows resource-constrained laptops to offload heavy model inference to a desktop GPU over a local network.

Limitations

• Official prebuilt releases target only macOS Sonoma and 64-bit Windows; Linux users must compile the Tauri and Rust source manually.