Microphone Setup
The #1 factor in transcription accuracy is audio quality.
Recommended External Microphones
Tested and ranked by the VORA team for meeting-room use. All prices are approximate USD retail (February 2026).
| Model | Type | Best For | Price | Verdict |
|---|---|---|---|---|
| Blue Yeti X | USB Condenser | Solo / podcast-style desk work | ~$130 | Top Pick |
| Rode NT-USB Mini | USB Condenser | Solo, compact desk setup | ~$100 | Best Value |
| Jabra Speak2 55 | USB Speakerphone | Small meeting rooms (2–4 people) | ~$200 | Conference Pick |
| Poly Sync 20 | USB Speakerphone | Small meeting rooms, portable | ~$150 | Conference Pick |
| Samson Go Mic | USB Clip-on | Budget, travel | ~$35 | Budget Pick |
| AirPods Pro 2 / WH-1000XM5 | Bluetooth Headset | Noisy offices, remote worker | $250–$350 | Headset Pick |
Placement & Room Tips
- Distance: Keep the mic 15–30 cm from your mouth. Too far = noise floor dominates; too close = plosives distort.
- Angle: Point a cardioid microphone directly at your mouth, not at the ceiling or screen.
- Room acoustics: Soft surfaces (curtains, carpets, bookshelves) absorb echo. A bare concrete room will produce noticeable reverb that confuses STT engines.
- Background noise: Close windows, mute fans if possible. Even a laptop cooling fan at 45 dB can drop accuracy by 5–15%.
- Bluetooth latency: Bluetooth mics add 100–200 ms of latency. VORA handles this gracefully, but wired USB is always preferable for accuracy.
In-App Audio Settings
Open Settings → Audio Processing in VORA:
- Noise Suppression (ON by default): Reduces background hum. Disable only if it causes artefacts on your mic.
- Echo Cancellation (ON by default): Essential if your speakers are near the mic. Disable only for headphone-only setups where there's no echo risk.
API Key Setup
VORA uses your own API keys — your data stays yours.
localStorage. They are never sent to VORA's servers.Gemini API Key (Primary AI)
Gemini handles AI question detection, meeting summaries, and text correction.
Go to Google AI Studio
Visit aistudio.google.com → click "Get API Key" → "Create API key".
Copy the key
The key starts with AIza.... Copy it immediately — it won't be shown again in full.
Paste into VORA
Open Settings (⚙) → Gemini AI → API Key field. Paste and click Save.
Verify the status
The indicator next to the field should turn green: "Gemini Ready".
Groq API Key (Fast Correction)
Groq runs Llama 3.3 70B on dedicated LPU hardware — blazing fast and completely free.
Create a free Groq account
Visit console.groq.com → sign up → go to API Keys.
Generate a key
Click "Create API Key". The key starts with gsk_....
Paste into VORA
Open Settings → Groq Fast Correction → API Key. Paste the key.
Enable Fast Mode
Toggle "Enable fast dual AI mode". VORA now sends every correction request to Groq first (≈300 ms) and falls back to Gemini for summaries.
OpenRouter API Key (Labs Only)
Used in the OpenRouter Parallel Lab to call three free LLMs simultaneously.
Create an OpenRouter account
Visit openrouter.ai → sign up → go to Keys.
Generate a key
Click "Create Key". The key starts with sk-or-....
Paste into the Lab
In the OpenRouter Parallel Lab page, enter your key in the API Key field at the top.
Meeting Context Settings
Context is the single biggest accuracy lever after microphone quality.
Why Context Matters
The AI corrector receives your speech transcript as raw text. Without context it must guess whether "프로포폴" is a drug name or a mishearing of "프로포절" (proposal). With the right context it gets it right every time.
VORA injects your context into every correction and summary prompt, behaving exactly like telling a human note-taker "this is a pharmaceutical regulatory meeting — watch for these terms."
Meeting Topic / Project Name
Found in Settings → Meeting Context → Topic / Project Name.
Write 1–3 sentences describing:
- What industry or domain the meeting is about
- The specific project or product being discussed
- Any key participants or roles
"Q2 regulatory review for Phase III clinical trial of compound XR-441. Participants: medical writers, regulatory affairs team. Key focus: FDA submission timeline and adverse event classification."
"Sprint planning for VORA v3.0 frontend refactor. Stack: React 19, TypeScript, Cloudflare Pages. Discuss: WebGPU rollout, Groq integration, A/B test results."
Priority Terms
Found in Settings → Meeting Context → Priority Terms.
Enter a comma-separated list of proper nouns, acronyms, product names, or technical terms that the AI should recognise and preserve exactly:
- Up to 30–40 terms work best. Too many dilute the signal.
- Mix Korean and English freely — the model handles both.
- Include common mis-transcriptions the system makes (e.g. "안드로이드" → "Android").
Meeting Category
Select the closest category in Settings → Meeting Category. This applies a pre-tuned persona to the AI:
- General — balanced correction, no domain bias
- IT / Software — tech terms, English-Korean code-switching
- Medical / Pharma — clinical terminology, drug names
- Legal — formal Korean / legal clauses
- Finance — financial jargon, numerical precision
- Marketing — brand names, campaign language
Cost & Plan Warnings
Understand your API costs before a long meeting.
Gemini API — Free vs Paid
Free Tier (no card)
- 15 requests / minute
- 1,500 requests / day
- 1 million tokens / minute
- Works well for short meetings (<30 min)
- Data may be used for model training
- Summaries may slow after 60 min
Pay-as-you-go ($5 minimum)
- 2,000 requests / minute
- Virtually unlimited daily quota
- Data not used for training
- SLA uptime guarantee
- Needed for multi-hour sessions
- ~$0.001–0.005 / correction call
Groq API — Free Tier Limits
| Model | Req / Min | Req / Day | Tokens / Min |
|---|---|---|---|
| Llama 3.3 70B Versatile (default) | 30 | 14,400 | 6,000 |
| Llama 3.1 8B Instant | 30 | 14,400 | 20,000 |
| Llama 3 8B 8192 | 30 | 14,400 | 20,000 |
For a typical 60-minute meeting with corrections every 3–5 seconds, VORA generates ~600–800 Groq requests — well within the 14,400/day limit.
OpenRouter — Free Model Limits
Free models on OpenRouter are subject to:
- 20 requests / minute per model
- 200 requests / day per model (with ≥$10 credits on your account)
- 50 requests / day per model (no credits / new account)
The OpenRouter Lab is intended for experimentation, not continuous 60-minute meetings. Use Groq for production correction and OpenRouter for side-by-side model comparison.
Estimated Monthly Cost Examples
| Usage Scenario | Gemini calls/mo | Est. cost |
|---|---|---|
| 2× weekly 30-min meetings, free tier | ~960 | $0 (free tier) |
| Daily 60-min meeting, free tier | ~18,000 | May hit daily limit |
| Daily 60-min meeting, paid tier | ~18,000 | ~$0.10–0.50 |
| Full workday (8 h/day), paid tier | ~144,000 | ~$1–4 |
Pro Tips & Troubleshooting
Common issues and quick fixes from the VORA team.
🎙️ Speech not being recognised?
- Check that your browser has microphone permission (click the lock icon in the URL bar).
- Open your OS audio settings and confirm the correct mic is set as default input.
- Try Chrome or Edge — the Web Speech API has the best support in Chromium browsers. Firefox and Safari have limited support.
- Reload the page and grant mic permission again if the indicator stays red.
🤖 AI correction not working?
- Open Settings and check the API status indicator — it should be green.
- If it shows "API key not set", re-enter your key and save.
- If it shows a 429 error, you've hit the rate limit. Wait 60 seconds or switch to Groq fast mode.
- Free tier Gemini keys from newly created Google accounts may be geo-restricted. Try creating the key on a desktop browser rather than mobile.
📝 Correction quality is poor?
- Fill in the Meeting Context and Priority Terms fields. This is the single biggest improvement lever.
- Select the correct Meeting Category (e.g., "Medical" for clinical meetings).
- Use the Language Selector to set the primary recognition language. Mixed Korean/English meetings should use
ko-KRas the primary language. - Reduce the Correction Interval to 2–3 seconds for fast-paced conversations; increase to 8–10 seconds for slower, deliberate speech to save API calls.
📊 Summary is missing or empty?
- Summaries require at least 50 words of transcript. Short meetings may not generate a summary automatically.
- Click Export Minutes manually to force a summary generation at any time.
- Ensure your Gemini key has remaining quota — Groq cannot generate summaries (correction only).
⌨️ Keyboard Shortcuts
| Shortcut | Action |
|---|---|
Alt + R | Start recording |
Alt + P | Pause / resume recording |
Alt + S | Stop recording |
Alt + E | Export meeting minutes |
Escape | Close settings modal |