Capture notes with high-accuracy multilingual voice input using OpenAI Speech-to-Text. Uses GPT-4o Audio Transcriptions (with tuned prompts for ja/en/zh/ko) and inserts into your notes with one click or push-to-talk.
- Voice activity detection modes: Off by default for maximum accuracy. Optional server-side chunking or local auto‑stop (requires fvad.wasm, installed manually)
Release assets: the plugin only needs `main.js`, `manifest.json`, and `styles.css` from the release bundle. Optional local VAD files are user-provided and are not included in releases.
Local VAD (optional): this plugin does not ship the WebAssembly file. If you want local VAD auto‑stop, download `fvad.wasm` from the fvad‑wasm project and place it in the same plugin folder, or use the “Choose fvad.wasm…” button in Settings → Voice Activity Detection to copy it. The plugin reads this local file from the installed plugin folder. No external network requests are made for local VAD. Local VAD is desktop‑only.
- **Transcription Language**: Japanese/English/Chinese/Korean (defaults to the UI language when linking is enabled; advanced settings let you override per language)
- Voice Activity Detection: choose Off (default, most accurate), Server (faster turnaround via upstream silence trimming), or Local (desktop auto‑stop; requires `fvad.wasm` installed manually)
- The plugin periodically auto‑saves your in‑progress text as a draft under `<vault>/.obsidian/plugins/voice-input/draft.txt` and restores it when you reopen the view.
- Drafts are device‑local and are not synchronized by Obsidian Sync or other services. If you clear the text area, the draft is removed.
- Processing in memory: audio is not written to disk by the plugin.
- Network use: recorded audio is sent over HTTPS to `https://api.openai.com/v1/audio/transcriptions` via Obsidian’s `requestUrl` for transcription. The settings connection test calls `https://api.openai.com/v1/models`. No telemetry, ads, or self-update requests are sent by the plugin.
- Account and billing: an OpenAI API key is required, and OpenAI API usage may be billed by OpenAI.
- API key storage: the API key is stored in plugin settings. When Electron SafeStorage is available, the key is encrypted with SafeStorage before saving.
- SafeStorage fallback: if Electron SafeStorage is unavailable, the plugin stores the key with a lightweight XOR/Base64 obfuscation fallback for backward compatibility. This fallback is not equivalent to OS-backed encryption.
- Clipboard access: the Copy button writes only the current transcription text to the system clipboard. If note creation or insertion fails, the plugin may also write that same transcription text to the clipboard as a recovery fallback. The plugin does not read from the clipboard.
- Vault file access: drafts are saved to, loaded from, and cleared from `<vault>/.obsidian/plugins/voice-input/draft.txt` using Obsidian Vault/FileManager APIs. Insert and append actions write only to the active target note, or create a timestamped `Voice-Memo-*.md` note when no suitable note is available.
- Local files: optional local VAD reads `fvad.wasm` from the plugin folder after you install or choose it. The release bundle does not include this WebAssembly file.
- External links: settings may show a link to the fvad-wasm GitHub project for manual download, but the plugin does not download those files automatically.
- Privacy policy: see [OpenAI’s Privacy Policy](https://openai.com/policies/privacy-policy/) for OpenAI API data handling.
