1. Choose live dictation or a saved recording
Live dictation converts what you are saying now into editable note text. A recording preserves audio so you can listen or transcribe it later. Decide which artifact belongs in the vault before you choose a microphone.
For daily notes, journaling, outlines, and rough Markdown, live text is usually the shorter path. For lectures, interviews, and meetings where playback matters, save audio and choose a separate transcription workflow deliberately.
2. Dictate into an Obsidian editing view
Open a note, switch to Live Preview or Source mode, click where the text should appear, then start the Dictation shortcut shown under System Settings → Keyboard. Apple’s system path produces text in the focused editable field; Reading view is not an editing surface.
Speak the prose first, then add Obsidian-specific Markdown such as links, properties, code fences, callouts, or complex lists with the keyboard unless your chosen speech tool explicitly supports and safely escapes those structures. Review headings and links before leaving the note.
3. Understand what Obsidian’s core Audio recorder saves
Obsidian documents Audio recorder as a core plugin that records audio, saves the file in the vault, and embeds it at the end of the active note. The recording remains in the vault even if you remove its embed from the note.
The official Audio recorder page does not document automatic speech-to-text. If you need a transcript rather than an audio attachment, use live dictation before the words enter the note or evaluate a separate transcription tool.
4. Treat community transcription plugins as privileged code
Obsidian’s security documentation says Restricted mode blocks community plugins by default. It also says Obsidian cannot reliably restrict plugins to specific permissions, so plugins inherit Obsidian’s access levels.
Before enabling a voice or transcription plugin, inspect its author, source, update history, requested API key, speech provider, retention policy, model downloads, and whether it can run locally. A plugin that writes into a local vault can still send audio or text to a remote service.
5. Separate local note storage from speech processing
Obsidian stores notes as Markdown-formatted plain text files in a vault on the local file system. That describes the note-storage layer, not the microphone or transcription layer.
Your complete privacy boundary can include the speech engine, formatting service, community plugin, sync provider, backups, and the vault location. A local vault plus a cloud transcription plugin is still a hybrid workflow; a local speech model plus a cloud-synced vault is another.
6. Diagnose failures without changing every layer
If system Dictation does not enter text, switch the note to Live Preview or Source mode and click the editor. Then test the same shortcut in TextEdit. If TextEdit also fails, check Keyboard → Dictation, the input source, language, and current processing requirement. If only Obsidian fails, try a plain new note with community plugins disabled.
For Obsidian’s Audio recorder or a plugin-owned microphone, check Obsidian under System Settings → Privacy & Security → Microphone. A working recording does not prove system Dictation insertion works, and working system Dictation does not prove a plugin has microphone access.
7. Use IraVoice for local capture and reviewable Markdown
IraVoice 0.7.1 is an option for macOS 26+ Apple-silicon Macs when speech recognition should run on the Mac or a rough technical thought should become reviewable Markdown before it reaches Obsidian. Models download on first use, Apple Intelligence is required for supported polishing and Voice-to-Spec, and Gumroad verifies paid activation.
IraVoice is not an Obsidian plugin and does not inspect or modify the vault. Use focused-app delivery where the editor exposes a supported field, or the clipboard when insertion is uncertain. For ordinary Markdown notes, review the generated structure before saving links, properties, or code.