Nubart Translate - Instructions

Critical Success Factor

Audio Quality Determines Everything

Nubart TRANSLATE requires:

Clean, isolated speech signal
No background voices
No feedback from translated audio
Consistent volume levels with no distortion
Clear voice signal without bass enhancement (which degrades AI transcription)

Internet Requirements

Network Compatibility:

Speaker's device must support secure WebSocket connections (WSS protocol). Most modern networks do; some corporate/public Wi-Fi blocks them.

Minimum Bandwidth (download speeds):

Event Size	Required Speed
Small (up to 50 users)	25 Mbps
Medium (50-200 users)	100 Mbps
Large (200+ users)	500 Mbps+

Ensure your router can handle simultaneous device connections.

How Nubart Optimizes Translation for Your Event

During purchase, we request:

Context documents (event agendas, technical papers, patents, manuals)
Speaker names
Which language(s) will be spoken during the event (not listener translation languages)
Any written materials related to the event subject

What we do with these materials:

AI-powered extraction: We use AI to analyze documents and extract a 3-page summary containing the most relevant technical terminology in natural language context
Multi-language preparation: We translate this context summary into your event's speakers' languages
Context feeding: The AI receives this contextual information (similar to briefing a human interpreter)
Internal testing: We read documents aloud through Nubart TRANSLATE to identify terms the AI doesn't recognize, then address them specifically

Why context documents work better than glossaries: AI-extracted context from real documents captures terminology in natural usage patterns. This produces better results than standalone term lists, which often contain common words AI already knows while missing truly unique terminology.

Audio Setup Requirements

Acoustic Isolation Required

The problem: Nubart TRANSLATE is highly sensitive to capture speech in large/noisy rooms. This means it captures ALL audio—including translated audio from listener devices.

We don't use a noise gate (would cut off speech and reduce accuracy).

Solution: Achieve ONE of these:

✅ All listeners use headphones/earphones
✅ Speaker and listeners in different locations
✅ Reduce mic sensitivity/room volume so mic captures speaker clearly but not room/listener devices

This is a physical acoustic requirement (same as professional interpreting booths).

Speaker Device

Recommended:

Laptop – Best stability and control
Recent smartphone – Fine if decent processor

Before event: Restart speaker device to close background processes.

Audio Input Options

Best to Worst Quality:

1️⃣ Professional Event Routing (STRONGLY RECOMMENDED)

Line/Aux output from PA/mixing console into speaker device, OR

2️⃣ Quality External Microphone

Headset, lavalier, or handheld mic
Provides hands-free, consistent audio

3️⃣ Built-in Device Microphone

Only acceptable in quiet environments
Vulnerable to noise, echo, room acoustics
Not recommended for important events

Your two main diagnostic tools:

1️⃣ Spectrum Display (1)

Shows audio signal like HiFi equipment
High tones right, bass tones left
Spectrum bars move when mic receives sound

Diagnostic Use:

Symptom	Cause
No movement when speaking	Mic not active or wrong device
Very small movement	Signal too weak, increase volume
Red bars across all frequencies	Overall signal too strong, reduce input volume
Red bars only in specific frequencies	Use equalizer to reduce those frequencies (lowering overall volume may weaken signal for transcription)

2️⃣ Audio Quality Indicator (5)

🟢 Green (80%+) - Good quality

🟡 Yellow (60-80%) - Usable, improve if possible

🔴 Red (below 60%) - Poor quality

If indicator shows red:

First: Verify correct spoken language selected
Check microphone is active and selected
Reduce background noise and echo
Ensure only speaker's voice reaches mic
Adjust microphone position/sensitivity

Target: Keep indicator green (80%+) throughout your event

Speaker & Listener Interfaces

Speaker Instructions

Before Starting (Day of Event)

Restart your speaker device for optimal performance.

1. Logging In

Go to www.nub.art/customer/login
Click "Nubart TRANSLATE"
If multiple rooms: Select your room

2. Select Spoken Language

Select the language you will speak during presentation.

⚠️ CRITICAL: If you change spoken language during event, ALWAYS use the "Change language" (4) before continuing. Forgetting this is the #1 cause of red quality indicators.

3. Starting Translation

Click "Unmute" (2)
Translation timer starts (appears in red)
Start speaking naturally—do NOT insert artificial pauses

Normal delay: First translated words appear a few seconds after you start speaking. Continue speaking naturally; system will catch up.

4. Monitor Audio Quality While Speaking

As you speak, verify these indicators:

Spectrum meter (1) shows visible movement when you talk
Audio Quality Indicator shows green (80%+)
No red bars in spectrum meter (1) (indicates distortion)

If any indicator is problematic:

Click "Mute" to pause translation
Fix the audio issue (adjust volume, change mic, reduce noise)
Click "Unmute" (2) again and verify indicators are now good

5. Pausing Translation

Click "Mute" (2)
Timer stops and turns black
System stops listening (stops billing)

6. Audience Questions (if enabled)

When listener submits question:

Red notification on "Questions" (3)
Click button to open question list
Questions automatically translated into your language
Recommended: Delete after answering

Listener Instructions

1. Joining the Translation

Scan the event QR code with smartphone
If multiple rooms: Select your room
Language selection: Default is your browser's language; change anytime using icon (6)

2. Reading and Listening

Text Display:

Translation appears as live text in center of screen
Adjust font size and brightness using icon (8)
Dark mode available to save battery

Audio Output:

By default, audio enabled (synthetic voice reads translation)
Mute/unmute anytime using icon (7)

If you can't hear audio:

Check device volume
Ensure audio not muted (7)
Close other apps using sound

3. Asking Questions (if enabled)

Tap "Questions" button (10) at bottom
Type question in language you're receiving translation
Speaker receives it translated into their language
Speaker decides when to answer

4. Exiting

Tap "Disconnect" (9) to exit translation.

Mandatory Pre-Event Testing

Why Testing Is Critical

5 minutes of testing prevents hours of poor translation.

Event organizers often skip testing because:

PA system sounds perfect
Recording quality is excellent
They assume AI "just works"

The Reality: Audio optimized for PA ≠ Audio optimized for speech-to-text. What sounds perfect through venue speakers may reach Nubart TRANSLATE with echo, distortion, or competing audio sources.

Testing reveals:

Actual audio quality reaching our system
Echo or feedback issues
Background noise problems
Microphone selection/volume issues
Network connectivity problems

The Test Procedure

Perform this 5-minute test at actual venue, under real event conditions, before your event begins.

Step-by-Step:

Log in as speaker on your main device (laptop/smartphone)
Use second device to scan event QR code
On second device:
- Select same language as speaker
- Mute audio output (7) to prevent feedback
Say 4-5 test sentences in presentation language
Check TWO indicators simultaneously:

Indicator A: Audio Quality (5)

On speaker device:

🟢 GREEN - Perfect! Continue.

🟡 YELLOW - Improve if possible

🔴 RED - Must improve

Must show GREEN while speaking.

Indicator B: Transcription Accuracy

On second device (listener view):

Read the live transcript in your language. Compare spoken words with displayed text.

Target: 90-95%+ match between spoken and displayed text

If significantly lower: Audio input quality is insufficient

If Test Results Are Poor

Problem	Solution
Red or yellow quality indicator	Verify correct language selected Improve microphone quality Reduce background noise
Transcription below 90%	Use external microphone Ensure only speaker's voice reaches mic Eliminate echo
Red bars in spectrum display	Lower input volume Increase mic distance
Spectrum display (1) barely moves	Increase input volume Move closer to mic
Wrong input device	Check OS settings Check browser settings Select correct microphone

Test Success Criteria

You're ready to go live when:

✅ Audio Quality Indicator (5) consistently shows GREEN (80%+)
✅ Transcription accuracy is 90-95%+
✅ Spectrum display (1) shows good signal without red bars
✅ No feedback loop (if both devices are in same room)

Common Testing Discoveries

Issue 1: Background Conversations

Symptom: Transcription includes words speaker didn't say

Cause: Microphone picking up nearby conversations

Solution: Use close-proximity mic (lavalier/headset), position speaker away from audience

Issue 2: Feedback Loop

Symptom: Translation repeats or stutters

Cause: Speaker's microphone hearing translated audio from listener devices

Solution: All listeners use headphones, or increase distance between devices

Issue 3: Incorrect Language

Symptom: Red quality indicator, nonsense transcription

Cause: Wrong spoken language selected (#1 cause of test failures)

Solution: Click "Change language" (4) and select correct language

Issue 4: Volume Too Low/High

Symptom: Spectrum display (1) barely moves OR shows red bars

Cause: Input volume incorrectly set

Solution: Adjust input volume in OS/browser settings, or adjust physical distance from mic

Pro Testing Tips

Test with real speaker: Have actual presenter perform test, not technician
Test at actual volume: Use same speaking volume as during event
Test with audience present: If possible, test with some attendees to verify acoustic isolation
Test all rooms: For multi-room events, test each room separately
Document settings: Note working mic settings, volume levels, and positioning
Test questions feature (3): If enabled, verify speaker can receive and see questions

When to Test

Ideal timing:

1-2 days before: Initial tech rehearsal
Morning of event: Final verification in actual conditions
15 minutes before start: Quick recheck

Don't skip testing just because you tested yesterday. Venue acoustics change with number of people present, and different technicians may alter audio settings.

Troubleshooting

Problem: "There is no translation"

Checks:

✅ Has speaker clicked "Unmute"?
✅ Does spectrum display show visible signal when speaker talks?
✅ Are both speaker and listener devices connected to internet?

If spectrum display shows very weak signal:

Input volume too low
Increase volume or move closer to microphone

Problem: "Translation doesn't make sense"

This is almost always due to poor audio input.

Immediate checks:

Audio Quality Indicator (5): Is it green? If yellow/red, improve audio quality
Correct spoken language? If not, use "Change language" (4)
Microphone picking up other sounds?
- Use headphones for all listeners
- Lower listener device volumes
- Increase distance between devices

Verification:

Run the audio input test (see Testing section) to check transcription accuracy.

Problem: "Audio Quality Indicator (5) shows red"

Two Most Common Causes:

1️⃣ Wrong Spoken Language Selected (50%+ of cases)

Solution: Click "Change language" (4) and select correct language

2️⃣ Poor Audio Input Reaching System

Symptom	Solution
Spectrum display (1) shows weak/no signal	Increase input volume
Spectrum display (1) shows red bars	Decrease input volume
Background noise/echo	Use better microphone, reduce room noise
Other voices being captured	Ensure only speaker's voice reaches mic

Problem: "Translation appears as text but I can't hear it"

Solutions:

Increase volume on listener device
Verify audio output enabled (7)
Close other apps using audio
Restart browser/app
Check device is not in silent/do-not-disturb mode

Problem: "Translation repeats sentences"

Cause:

Speaker's microphone is hearing the translated audio output.

Common Scenario:

During testing, one person's devices are too close together.

Solutions:

Mute audio on listener device
Use headphones
Move devices farther apart
Reduce listener device volume

Problem: "Translation is very delayed"

Normal Delay:

Up to 2 seconds after speaker finishes a sentence is normal (similar to human interpreters).

If delay is significantly longer:

Check internet connection on both speaker and listener devices
Verify bandwidth is sufficient for event size
Close unnecessary browser tabs/apps
Move closer to Wi-Fi router

Problem: "Translation takes too long to begin"

Normal behavior: AI needs a few seconds to begin processing. It's normal for first translated words to appear a few seconds after speaker starts.

Speaker should NOT:

❌ Wait for translation to catch up
❌ Insert artificial pauses
❌ Speak slower than natural pace

Speaker SHOULD:

✅ Continue speaking naturally
✅ System will catch up automatically

Problem: "Questions not appearing for speaker"

Checks:

Is questions feature enabled? (Contact us if unsure)
Has listener actually submitted a question?
Is speaker looking at "Questions" (3) for red notification?
Has speaker clicked the Questions button to open list?

If using separate device for questions:

Verify device is logged in with correct account
Ensure device is connected to internet
Check microphone on this device is MUTED

Device Issues

Symptom: Browser/app crashes or freezes

Solutions:

Restart browser/app
Clear browser cache and cookies
Update browser to latest version
Try different browser (Chrome, Firefox, Safari)
Restart device
Close other apps consuming resources

Microphone Issues

Symptom: No audio input detected

Checks:

Is microphone physically connected?
Is correct input device selected in OS settings?
Is correct input device selected in browser settings?
Is microphone muted in OS/browser?
Does microphone need batteries/charging?
Is another app using the microphone?
Has browser been granted microphone permissions?

Testing Microphone:

Close Nubart TRANSLATE
Test mic in OS settings (Windows Sound Settings, Mac System Preferences, etc.)
Test mic in browser (Chrome > Settings > Privacy and Security > Microphone)
If working in tests, reopen Nubart TRANSLATE

When to Contact Support

Contact us if:

Testing reveals persistent red quality indicator (5) despite following all troubleshooting steps
You want to change the colours or logo of the listener interface
You need help with multi-room event setup
You need help with videoconference integration
Questions feature not working
Billing/usage questions

Contact Information:

Email: info@nubart.eu

Response Times:

Business hours (9:00-18:00 CET/CEST, Mon-Fri): Within 24 hours
Emergency during confirmed events: Contact info provided after purchase

When Contacting Support, Include:

Event date and time
Number of expected attendees
Description of issue
Audio quality indicator (5) reading
Screenshot of interface (if possible)
Steps already taken to troubleshoot

Special Scenarios

Videoconference Translation

For remote speakers, route Zoom/Meet audio from device A to Nubart on device B via a TRRS cable. Test this setup days in advance. For technical details and setup instructions, please refer to the FAQ on our website.

Multi-Room Events

Before Event:

Inform us of:

Number of rooms
Room names (e.g., "Room 1," "Venus Room")
If rooms cover different topics: Brief description for each (helps AI context)

Speaker Account Setup (Multi-Room)

Each room requires separate speaker account with different email:

Log in to www.nub.art/customer/login
Go to "Employees" tab
Click "Invite employee" for each room
Enter name and email for each
IMPORTANT: Select "Guide and Translate" role
System sends invitation to each email

Note: You need 'admin' permissions to access Employees tab. Contact us if you don't see this option.

On Event Day:

Speakers:

Log in and see list of all rooms
Select correct room

Listeners:

Scan QR code
See list of all rooms
Select room they're attending

QR Code Strategy:

Single QR for all rooms: Simplifies logistics (recommended)
Separate QR per room: Only if rooms very far apart

⏱️ Billing Note: Translation time counted per room. One hour across 3 simultaneous rooms = 3 billable hours.

Speaker Device Out of Reach

Scenario: Device sending audio is not where speaker can see it

Example: Laptop connected to sound system is in a booth or back-of-room, but speaker is on stage.

Problem:

Speaker can't see incoming audience questions.

Solution: Create Second "Guide and Translate" Account

Setup steps:

Go to "Employees" tab in Customer Area
Click "Invite employee"
Assign a name (e.g., "Speaker Monitor - Room A") and email
Complete invitation process to activate account
Log in to new account on second device (tablet/phone)
Open the Questions panel (3) so questions are visible as they arrive
Place this device in speaker's view (on podium/table)

⚠️ CRITICAL: Keep microphone MUTED on the second device at all times!

What speaker sees on second device:

Incoming audience questions (translated into speaker's language)
Questions appear automatically as listeners submit them

What speaker can do:

Read questions as they arrive (no touching required)
Remove questions already answered (click X on question)

Primary device responsibilities (sound technician in booth):

Handles audio transmission
Monitors Audio Quality Indicator (keep green)
Adjusts audio settings if quality drops

Note: If you lack admin permissions to create employee accounts, contact Nubart support at info@nubart.eu. We can also disable audience questions completely.

Panel Discussions / Multiple Speakers

Scenario: Multiple speakers sharing one microphone

Option 1: One speaker account (simplest)

One person controls "Unmute/Mute" (2)
All panelists share microphone
Works if all speak same language

Option 2: Multiple speaker accounts

Each panelist has own account
Each unmutes (2) when their turn
Allows different source languages
Requires coordination between speakers

Tip: For panels, designate one "technical operator" to control unmute/mute timing. This prevents confusion and overlapping audio.

Multi-Day Conferences

Day 1 Recommendations:

Arrive 1 hour early for thorough testing
Test each room/speaker separately
Document working settings for each room
Identify any problematic speakers/locations

Day 2+ Recommendations:

Quick 15-minute retest before start
Verify no overnight changes to audio setup
Check different technicians haven't altered settings

Speaker Rotation:

Brief each new speaker on "Unmute" (2) and "Change language" (4) buttons
Emphasize importance of clicking "Mute" when finished
Have technical operator on standby for first-time speakers

International Events / Multiple Languages

If speakers present in different languages:

Each speaker MUST:

Click "Change language" (4) before starting
Select their spoken language
Verify green quality indicator (5)

⚠️ Critical: Forgetting to change spoken language is the #1 cause of "nonsense translation" complaints. Brief all speakers on this requirement.

Pre-Event Speaker Briefing:

Create a simple one-page guide for speakers covering:

How to log in
How to select their spoken language
How to click Unmute (2)
How to click Mute when finished
Where to look for audience questions

Consider providing this in multiple languages for international speakers.

What to Expect from AI Translation

Professional Excellence, Not Perfection

Nubart TRANSLATE provides professional-grade simultaneous interpretation comparable to experienced human interpreters (85-95% accuracy range).

AI translation is NOT:

Mathematical certainty (language is subjective)
Perfect like document translation (DeepL/ChatGPT)
Infinitely customizable through glossaries

Simultaneous ≠ Document Translation

Document Translation (DeepL, ChatGPT)	Simultaneous Interpretation (Nubart TRANSLATE)
Complete texts, knows sentence endings	Live, doesn't know how sentence continues
Perfect text input	Must transcribe speech first (accents, noise)
No time pressure	Immediate delivery required

Double Challenge: AI must first convert speech to text (with accents, noise, imperfect mics), THEN translate. The analogy: Comparing simultaneous interpretation with document translation is like comparing a live broadcast with a studio-edited video. Both can be excellent, but they play by completely different rules.

How Nubart Helps: We analyze your event's context documents (agendas, technical papers, manuals) using AI to extract relevant terminology in natural language. This context is translated and fed to the system—like briefing a human interpreter. See Page 1 for details.

AI vs. Human Interpreters

AI Advantages	AI Challenges
✅ Superhuman memory (terminology, figures)	❌ Background voice separation
✅ Never fatigued	❌ Too precise (replicates errors)
✅ Instant availability	❌ Punctuation placement
✅ Unlimited scalability	❌ Doesn't smooth speech
✅ Fraction of cost	❌ Language mixing (code-switching)
	❌ Foreign name pronunciation (TTS)

Three Common Limitations

1. Language Mixing (Code-Switching)

Scenario: German speaker uses English terms ("customer journey," "machine learning").

Problem: AI set to German doesn't know if these are loan words (keep in English) or need translation.

Solution: If you know speakers will use specific English terms, mention these in your context documents. Brief speakers to stay consistent; use "Change language" button for full sections.

2. TTS Pronunciation of Foreign Names

Example: German "Fraunhofer" → Text ✅ correct → English TTS voice ❌ "Frown-hoffer"

May affect: Company names (BMW, Siemens), people (François, Müller), places (München, København)

Why: TTS uses target language phonetic rules, doesn't auto-switch for foreign names.

Solution: Inform listeners text is accurate; recommend text-only mode for events with many foreign names.

3. External Conditions Impact Quality

Factor	Impact
Audio clarity/volume	🔴 High
Environmental noise	🔴 High
Room acoustics	🔴 High
Speaker pace/articulation	🟡 Medium
Internet quality	🟡 Medium
Speaker accent	🟡 Low-Med

AI is not magic. Requires optimal input conditions like human interpreters need good equipment and booth isolation.

When AI Works Best

✅ Technical presentations (specialized vocabulary)
✅ Business conferences (standard terminology)
✅ Academic lectures (well-defined fields)
✅ Product demos (clear structure)
✅ Panel discussions (prepared speakers)

Challenging Scenarios

❌ Highly colloquial language (idioms/slang)
❌ Rapid-fire dialogue (interruptions)
❌ Poetry or wordplay
❌ Poor audio (echo, noise, low volume)
❌ Language mixing within sentences

Note: Human interpreters also struggle with these—not AI-specific.

Setting Attendee Expectations

Communicate in pre-event materials:

✅ "AI-powered real-time translation available"
✅ "Quality comparable to professional interpreters"
✅ "Bring headphones for best experience"
✅ "35+ languages available"

Don't promise:

❌ "100% accurate translation"
❌ "Perfect voice reproduction"
❌ "Zero delay"

The Revolution

Technology has democratized professional interpretation. Events of any size can offer multilingual access at a fraction of traditional cost.

AI improves rapidly:

Better accent recognition
Improved contextual understanding
More natural voice synthesis
Faster processing, lower latency

What seems challenging today may work flawlessly next year.

Final Checklist & Contact

Final Checklist Before Your Event

✅ 48-72 Hours Before:

[ ] Internet bandwidth verified for expected attendance
[ ] Network supports WebSocket (WSS) connections
[ ] Speaker accounts created for each room (if multi-room)
[ ] QR codes printed and distributed
[ ] Professional audio routing arranged (Line/Aux from PA, or Y-splitter)
[ ] Speakers briefed on basic operation

✅ On-Site Before Event:

[ ] Speakers' devices restarted
[ ] Audio input test completed
[ ] Audio Quality Indicator shows green when speaking
[ ] Transcription accuracy verified at 90%+
[ ] Spectrum display shows good signal without red bars
[ ] Acoustic isolation confirmed (listeners using headphones, or proper separation)
[ ] Questions feature tested (if enabled)
[ ] Multi-room selection tested (if applicable)
[ ] Backup audio equipment available

✅ During Event:

[ ] Monitor Audio Quality Indicator—must stay green
[ ] Speaker uses "Change language" if switching languages
[ ] Timer running (red) when unmuted, stopped (black) when muted
[ ] Technical operator monitoring for issues
[ ] Backup contact for support available

Common Mistakes to Avoid

Skipping pre-event testing
→ Test at venue under real conditions
Using built-in device microphone
→ Use professional audio routing or external mic
Forgetting to change spoken language
→ Always use "Change language" button (4)
Not using headphones for listeners
→ Causes feedback loops and poor audio
Submitting 100+ term glossary
→ Provide good written context information
Ignoring Audio Quality Indicator
→ Must stay green throughout event
Expecting mathematical perfection from AI
→ AI provides professional excellence, not perfection
Not briefing speakers on basic operation
→ Create simple one-page guide for speakers

Contact & Support

Email:

info@nubart.eu

Response Times:

Business hours (9:00-18:00 CET/CEST, Monday-Friday):
Within 24 hours
Emergency support during confirmed events:
Contact information provided after purchase

When to Contact Support:

Testing reveals persistent issues despite troubleshooting
Multi-room event setup assistance
Videoconference integration help
Customization requests
Billing or usage questions

Include in Support Requests:

Event date and time
Expected number of attendees
Description of issue
Audio quality indicator reading (if applicable)
Screenshots (if possible)
Troubleshooting steps already taken