I built an open‑source alternative to Typeless that can also run fully offline

Description and use

The author built an offline, privacy‑focused voice‑to‑text tool that replaces Typeless, supporting multiple ASR models, local processing, and secure mobile microphone integration.

Problem

Users need a voice‑to‑text tool that works offline, respects privacy, and can be used in any application without a wake word or continuous listening.

Outcome

The author delivered a fully offline, privacy‑preserving voice‑to‑text tool that can be used in any application via a keyboard shortcut, with optional AI polish and dictionary replacements. It supports multiple ASR models, GPU acceleration, and mobile microphone integration.

Workflow

• 1. User presses Ctrl+Alt+Space to start recording. 2. Audio is captured only while the shortcut is held. 3. Recognition runs locally using one of five downloadable models. 4. Dictionary replacements are applied. 5. Optional AI polish refines the text. 6. Replacement rules adjust the output. 7. The final text is pasted at the cursor and the original clipboard is restored.
• 8. For mobile use, a phone can act as the computer’s microphone via a secure Opus stream.
• 9. A headless Linux server (voltip‑server) can provide the same service with an OpenAI‑compatible API.
• 10. All text sent to external AI services is pre‑processed; no audio is transmitted.
• 11. API keys are stored in the system keychain and sent only to their respective providers.
• 12. The tool supports offline operation and can be run on GPU or CPU depending on the model and platform.

Evidence

Hold Ctrl+Alt+Space, speak, let go; the text is pasted at the cursor in any app, then the clipboard is put back.

Source

Five downloadable models: Qwen3-ASR 0.6B (690 MB, 30 languages, punctuation, GPU or CPU), Qwen3-ASR 1.7B (1.7 GB), SenseVoice Small (240 MB, CPU), Paraformer (227 MB, no punctuation, CPU), live transcription (169 MB, preview only).

Source

Your numbers (16 s of Chinese, Qwen3-ASR 0.6B, after the first run): L40S Vulkan 0.16 s, same machine CPU 8 threads 2.87 s, Tesla T4 Vulkan 0.28 s.

Source

Pairing: QR code or 6-digit code, both screens show the same safety code.

Source

Text only: recognised text after the dictionary, the preset's instructions, the dictionary's right spellings, the app name (can be turned off). Never audio.

Source