Audio recorder

Audio recorder

The audio recorder mirrors Obsidian's core audio recorder: start recording from a note, stop it, then obsidian.nvim attaches the audio file to the vault and inserts the attachment link at the cursor position that started the recording.

Code actions

Use your normal LSP code action mapping in a note:

• Start recording audio as attachment starts recording.
• The same action changes to Stop recording audio while a recording is active.

The insertion point is tracked with an extmark, so edits above the cursor while recording should not move the final link to the wrong place.

Recording backend

By default obsidian.nvim looks for one of:

• rec from sox
• sox
• arecord

:checkhealth obsidian reports the available backend. Starting a recording also warns if no backend is available.

The temporary recording file uses a .wav suffix because the built-in CLI backends write WAV/PCM audio. Obsidian's app commonly records .m4a, but these CLI tools do not reliably encode m4a without extra codec support.

Storage

Recordings are first written to a temp file and then copied with obsidian.attachment.add() into your configured attachments folder using an Obsidian-style Recording YYYYMMDDHHMMSS.wav name. The temporary file is deleted by default after it is attached.

obsidian.nvim logs both paths when the recording is attached.

Callback for transcription or summaries

Stopping a recording adds the audio through the normal attachment pipeline. Use callbacks.add_attachment or the ObsidianAttachmentAdded user autocmd to run transcription, summary, or other custom logic after the link is inserted.

Minimal Whisper API example

This callback sends recorded audio files to OpenAI's Whisper API and appends the transcript below the note:

local function transcribe_with_whisper(path, ctx)
  local key = vim.env.OPENAI_API_KEY
  if not key or key == "" then
    vim.notify("OPENAI_API_KEY is not set", vim.log.levels.ERROR)
    return
  end

  vim.system({
    "curl",
    "-sS",
    "https://api.openai.com/v1/audio/transcriptions",
    "-H",
    "Authorization: Bearer " .. key,
    "-F",
    "model=whisper-1",
    "-F",
    "file=@" .. tostring(path),
  }, { text = true }, function(obj)
    vim.schedule(function()
      if obj.code ~= 0 then
        vim.notify(obj.stderr or obj.stdout, vim.log.levels.ERROR)
        return
      end

      local ok, decoded = pcall(vim.json.decode, obj.stdout)
      if not ok or not decoded.text then
        vim.notify("Failed to parse Whisper response", vim.log.levels.ERROR)
        return
      end

      vim.api.nvim_buf_set_lines(ctx.buffer, -1, -1, false, {
        "",
        "## Transcript",
        "",
        decoded.text,
      })
    end)
  end)
end

require("obsidian").setup {
  callbacks = {
    add_attachment = function(path, ctx)
      if ctx.scope == "audio_recorder" then
        transcribe_with_whisper(path, ctx)
      end
    end,
  },
}