Deepgram Meeting STT

Deepgram Meeting STT

An Obsidian plugin that turns meeting recordings into Markdown notes — speech-to-text via Deepgram with per-speaker timestamps, plus an optional AI summary step via Google Gemini that drops a structured digest next to the transcript.

🇰🇷 한국어 가이드: README-ko.md

---

What you get

• Audio → transcript: drop an audio file, right-click, pick a title — get back a Markdown note with per-speaker [HH:MM:SS] timestamps.
• Audio → transcript → AI summary (optional): in the same right-click, also produce a structured summary note (action items, decisions, key quotes — whatever the template says) backlinked to the transcript.
• File-based summary templates under ObsiDeep/Templates/. Each .md declares its own prompt and output skeleton with {{placeholders}}; you edit them like any other note. Favorites surface as flat menu items; the rest sit in an AI 요약 ▸ submenu. Three starters are seeded on first run (Meeting, Interview, Lecture).
• Workspace stays clean: ObsiDeep/ (Audio · STT · Templates · AI-Summaries) is auto-created at the vault root, and .gitignore rules are added so recordings and your API keys never enter vault git sync.
• Bilingual UI — Korean / English / auto-follow Obsidian locale.
• Zero Retention by default for Deepgram; AI summary is fully optional and gated on having a Gemini key set.
• Mobile compatible — works on Obsidian for iOS / Android.

The plugin keeps working as a pure STT tool. If you don't set a Gemini key, no AI menus appear and nothing leaves your machine except the audio you choose to transcribe.

Install

1. Settings → Community plugins → Browse
2. Search for "Deepgram Meeting STT"
3. Install + Enable

Setup

On first enable, a one-time consent modal explains what's sent to Deepgram and auto-creates an ObsiDeep/ workspace at the vault root — including Templates/ (seeded with three starter summary templates) and AI-Summaries/. It also writes .gitignore rules so recordings and your API keys stay out of vault git sync.

1. Deepgram API key (required for STT)

1. Sign up at Deepgram Console — free tier includes a $200 credit (~770 hours of nova-3).

   

2. API Keys → Create New API Key (any name, Member permission is enough).

   

3. Paste into Settings → Deepgram Meeting STT → "Deepgram API key" → click Validate.

   

2. Gemini API key (optional — only for AI summary)

Leave this blank if you only want STT. To enable AI summary:

1. Open Google AI Studio → API keys and create one.
2. Paste into Settings → Deepgram Meeting STT → "Gemini API key".
3. Pick a model — gemini-2.5-flash (default, fast & cheap) or gemini-2.5-pro (higher quality).

The AI menus stay hidden until a key is set.

Usage

Transcribe only

1. Place an audio file (mp3, m4a, mp4, wav, flac, ogg, opus, webm, aac) into ObsiDeep/Audio/.
2. Right-click the file → ObsiDeep ▸ STT만 추출 (Transcribe only).
3. Enter a title → Enter.
4. After ~1–2 minutes the transcribed note appears in ObsiDeep/STT/ and opens automatically.

The same flow is available from the command palette (Cmd+P / Ctrl+P) → "Transcribe audio → meeting note".

If you copied the file via Finder/Explorer and it doesn't appear in the sidebar right away, run Cmd+P → "Reload app without saving" first.

Transcribe + summarize in one go (AI summary set up)

1. Right-click an audio file → ObsiDeep ▸ ⭐ STT + 요약: Meeting (favorite templates surface flat; others live under AI 요약 ▸).
2. Enter a title.
3. After STT finishes, the transcript is fed to Gemini with the template's prompt.
4. Two notes appear: the transcript in ObsiDeep/STT/, the summary in ObsiDeep/AI-Summaries/{title} (요약).md with a source: "[[...]]" backlink to the transcript.

If STT fails, nothing else runs. If only the summary step fails, the transcript is preserved and you get a notice.

Re-summarize an existing note

Right-click any markdown note → ObsiDeep ▸ AI 요약: <template> — same flow, but the note's body is the input transcript. Useful for trying a different template, or summarizing notes that weren't produced by this plugin.

Output examples

Transcript (ObsiDeep/STT/...):

---
date: 2026-05-13
type: meeting
tags: [meeting, stt]
duration: 28m 41s
language: ko
source: [[ObsiDeep/Audio/standup.m4a]]
speakers: ["Speaker 1", "Speaker 2"]
---

# Stand-up 2026-05-13

**Speaker 1** [00:00:01 - 00:00:08]
Good morning, let's start with the status updates.

**Speaker 2** [00:00:09 - 00:00:14]
Sure, I'll go first.

Summary (ObsiDeep/AI-Summaries/..., using the Meeting template):

---
source: "[[ObsiDeep/STT/Stand-up 2026-05-13]]"
template: "Meeting"
date: 2026-05-13
language: Korean
type: ai-summary
---
# Stand-up 2026-05-13

> Source: [[ObsiDeep/STT/Stand-up 2026-05-13]] · 2026-05-13 · 28:41

## Summary
- Status updates from each team
- ...

## Decisions
- ...

## Action Items
- [ ] @Speaker1 — follow up with ...

Customize a template

Edit any .md in ObsiDeep/Templates/:

---
name: "Meeting"
favorite: true        # show flat in the right-click menu
prompt: |
  Summarize the transcript as meeting minutes.
  - decisions: clear, one bullet per decision.
  - action_items: use checkbox markdown; include owner name when present.
placeholders:
  summary: "3-5 bullet overview of the discussion"
  decisions: "Bullet list of decisions made"
  action_items: "Checkbox list of action items (with owner if mentioned)"
---
# {{title}}

> Source: {{source}} · {{date}} · {{duration}}

## Summary
{{summary}}

## Decisions
{{decisions}}

## Action Items
{{action_items}}

The placeholders you declare here become the JSON schema Gemini is forced to fill — no free-form text drift. Add as many as you want.

To scaffold a new template with every system placeholder documented, run the command palette → "새 요약 템플릿 만들기 (Create new summary template)".

Rename speakers

Speakers are labelled Speaker 1, Speaker 2 by default. To replace with real names:

1. Open the meeting note.

2. Command palette → "Rename speaker (current note)".

   

3. Pick the speaker from the dropdown (auto-detected from the note's frontmatter).

4. Type the real name → click Replace.

Every occurrence in both the body and the speakers frontmatter array is rewritten.

Settings

See FEATURES.md for the full placeholder reference, AI summary deep-dive, accuracy guide (audio quality, speaker count, recording-room checklist), and mobile usage notes.

Security & Privacy

• Audio is sent to Deepgram over HTTPS for processing. Transcript text is sent to Gemini only if you trigger an AI summary.
• Your API keys are stored locally as plain JSON in .obsidian/plugins/deepgram-meeting-stt/data.json (Obsidian plugin standard). The plugin auto-adds this path to your vault's .gitignore.
• The ObsiDeep/ folder is also auto-added to .gitignore so recordings, transcripts, and summaries never enter vault git sync.
• Deepgram complies with GDPR / SOC 2 Type II / HIPAA (with BAA) / CCPA. See Deepgram Trust Center and Privacy Policy.
• Zero Retention is on by default for Deepgram. Guaranteed immediate effect on Growth or higher Deepgram plans; free / Pay-as-you-go tier may still retain data per standard policy (~30 days).
• Gemini API usage is governed by Google's Gemini API Terms and Privacy Policy. Free-tier requests may be used to improve Google's models — use the paid tier if that's a concern.
• Please obtain consent from meeting participants before recording and transmitting audio to a third-party API.

Cost

Approximate cost with nova-3:

The free $200 Deepgram credit + Gemini's free tier cover many hours of typical use. See Deepgram Pricing and Gemini API Pricing for current rates.

License

MIT © 2026 Moonjuun