The latest articles on DEV Community by sunerpy (@rain_charles_3fcf80f35d59). https://dev.to/rain_charles_3fcf80f35d59 https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4181526%2Fa1e217e3-5cef-4a5c-8203-955707dea38a.png https://dev.to/rain_charles_3fcf80f35d59 en sunerpy Thu, 22 Oct 2026 09:30:15 +0000 https://dev.to/rain_charles_3fcf80f35d59/i-built-an-open-source-alternative-to-typeless-that-can-also-run-fully-offline-276b https://dev.to/rain_charles_3fcf80f35d59/i-built-an-open-source-alternative-to-typeless-that-can-also-run-fully-offline-276b <h2> What happens between "hold the key" and "text at the cursor" </h2> <ul> <li>Hold Ctrl+Alt+Space, speak, let go; the text is pasted at the cursor in any app, then the clipboard is put back. Not an input method; the microphone is open only while the shortcut is held. No wake word, nothing spoken gets executed.</li> <li>The order: recognition → dictionary replacements → AI polish (optional) → replacement rules → paste.</li> <li>(Your example: "Um, the review is on Monday at ten. No, sorry, at two…" → "The review is on Monday at two. Please bring the latest numbers." The gif shows it.)</li> </ul> <h2> Recognition on your own machine </h2> <ul> <li>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).</li> <li>Downloads: project mirror → Hugging Face → hf-mirror.com; resumable; disk space checked; every file verified against its SHA-256. Offline import via "Check and import".</li> <li>GPU only for Qwen3-ASR: Vulkan on Windows/Linux, Metal on macOS; falls back to the CPU.</li> <li>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. First GPU run is slow (~13 s on the L40S, ~2 min on the T4), so the model is loaded in the background at start.</li> <li>x86 needs AVX2 for Qwen3-ASR. AMD/Intel GPU measurements are still on the roadmap (ask readers for numbers).</li> </ul> <h2> The phone as the computer's microphone </h2> <ul> <li>Pairing: QR code or 6-digit code, both screens show the same safety code.</li> <li>Audio: Opus, end-to-end encrypted, streamed through a relay that cannot read it; the computer recognises with its own settings. The phone runs no local model.</li> </ul> <h2> voltip-server </h2> <ul> <li>Same recognition, dictionary and presets on a headless Linux server; OpenAI-compatible transcription API; accepts connections from that machine only unless opened up.</li> </ul> <h2> What the AI service receives </h2> <ul> <li>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.</li> <li>Keys in the system keychain; each key sent only to its own provider. Fallback models (up to 8) only on an explicit quota-exhausted answer.</li> </ul> <h2> Trade-offs (why there is no wake word / silence auto-stop, admin windows on Windows, </h2> <p>unsigned Windows installer, macOS not notarized, iOS planned)</p> <h2> Links </h2> <ul> <li><a href="https://github.com/sunerpy/voltip" rel="noopener noreferrer">https://github.com/sunerpy/voltip</a></li> <li><a href="https://voltip.firlab.app/guide/try" rel="noopener noreferrer">https://voltip.firlab.app/guide/try</a></li> </ul> dictation opensource