Live Transcription

UniFlow's Live Transcription captures audio in real time and converts it to text using Deepgram's Nova-2 speech recognition engine. Whether you're in a lecture hall, a Zoom call, or a study group, every word is transcribed, timestamped, and automatically indexed into UniMind's knowledge graph — making every lecture searchable and citable.

Plan requirement: Live Transcription is available on Starter, Pro, and Max plans. Free users can purchase Transcription Time Packs for on-demand access.

Getting Started

1. Click Live Transcription in the sidebar (or the microphone icon in the title bar)
2. Select the course this session belongs to (optional but recommended for AI context)
3. Click Start Recording
4. Grant microphone permission when prompted
5. Begin speaking or playing audio — text appears in real time

How It Works

Audio Pipeline

1. Microphone capture — UniFlow uses the Web Audio API with an AudioWorklet processor for low-latency, high-fidelity audio capture at 24 kHz
2. Noise processing — Configurable input gain, noise suppression, echo cancellation, and auto gain control
3. Streaming to Deepgram — Raw PCM audio is streamed over WebSocket to Deepgram's Nova-2 model
4. Real-time results — Interim (partial) results appear instantly; final results are confirmed when each sentence completes
5. Utterance detection — Deepgram's UtteranceEnd event detects natural pauses between speakers or topics

Transcription Display

The transcription notebook shows text in a clean, scrollable interface:

• Interim text appears in a lighter style as you speak, updating in real time
• Finalized text is locked in once Deepgram confirms the sentence
• Automatic segmentation groups sentences into paragraphs (every 5 sentences by default)
• Timestamps are recorded for every segment

Key Features

Bilingual Support

UniFlow supports transcription in multiple languages:

• English (default) — Optimized for academic lectures
• Chinese — Full Mandarin support
• Automatic language detection based on your settings

Microphone Sensitivity Profiles

Choose the right profile for your environment:

Configure this in Settings > General > Microphone Sensitivity.

Recording Controls

• Pause / Resume — Temporarily pause without ending the session
• Stop — End the recording and finalize the transcript
• Auto-recovery — If the microphone stream is lost (e.g., device disconnected), UniFlow automatically attempts to reconnect

Session Management

Each transcription is saved as a session with:

• Session title (editable)
• Associated course
• Creation date and duration
• Full transcript with timestamps
• Recording status (recording, paused, completed)

You can browse all sessions in the Transcription Library, filter by course, and search across transcripts.

Classroom Q&A Mode

One of UniFlow's most powerful features is the Classroom Q&A Mode, accessible via the Mini Chat panel during a live transcription session.

What It Does

When enabled, the AI reads your live transcription in real time and helps you:

• Identify the professor's latest question from the transcript
• Generate a ready-to-speak answer in bilingual format (English + Chinese)
• Prepare classroom contributions with the right tone — confident but not overly formal

How to Use It

1. Start a transcription session
2. Click the chat icon to open the Mini Chat drawer
3. Toggle Classroom Q&A Mode on
4. Use quick actions like:
   • "Help me answer" — Finds the latest question and generates a response
   • "Class participation" — Prepares a discussion contribution
   • "Add a point" — Builds on the current discussion
   • "Summarize discussion" — Summarizes key points with your position
   • "Ask a question" — Prepares a thoughtful question to ask
   • "Respond to classmate" — Crafts a response to another student's point

Answer Format

All Classroom Q&A responses follow a strict bilingual format:

**English:**
I think the main factor here is... because...

**中文：**
我认为这里的主要因素是……因为……

The tone is calibrated to sound like a well-prepared student speaking in class — clear, structured, and natural.

Normal Mode

When Classroom Q&A Mode is off, the Mini Chat works as a general AI assistant with access to the live transcript. You can ask:

• "What's being discussed right now?"
• "Are there any tasks assigned?"
• "Explain that last part"
• "Summarize this lecture"
• "Highlight the key points"
• "Generate study notes"

Cloud Sync

Transcription sessions are automatically synced to the cloud:

• Access your transcripts on any device
• Transcribed content is indexed into UniMind's knowledge graph
• Every word from every lecture becomes searchable and citable when you use UniMind

Transcription Time & Billing

Transcription Time Packs

Time Packs are one-time purchases that add extra high-precision transcription time to your account:

• Work with any plan (including Free)
• Valid for 180 days from purchase
• Stack with your subscription's included time
• Purchase from the Pricing page

Tips

• Associate sessions with courses — This enables UniMind to use lecture content when answering questions about that course
• Use the Sensitive profile in large halls — It captures distant audio better by reducing noise suppression
• Keep the app in focus — Some browsers throttle background audio processing; keeping UniFlow in the foreground ensures the best capture quality
• Review and edit after class — Transcripts can be edited for accuracy before they're indexed

Troubleshooting

Microphone not detected

1. Check that your microphone is connected and not muted
2. Ensure UniFlow has microphone permission in your system settings
3. Try selecting a different audio input device in your system preferences
4. Restart the app

Transcription quality is poor

• Switch to the Sensitive microphone profile for distant speakers
• Ensure you're in a relatively quiet environment
• Check your internet connection — Deepgram requires a stable WebSocket connection
• Try moving closer to the audio source

Connection drops during recording

UniFlow includes automatic reconnection logic. If the Deepgram WebSocket disconnects:

1. Audio chunks are buffered locally (up to 100 chunks)
2. The client attempts to reconnect automatically
3. Buffered audio is replayed once the connection is restored

If reconnection fails, you'll see an error indicator. Try stopping and restarting the session.

What's Next

• UniMind Engine — Ask AI questions using your lecture transcripts as context
• Canvas Sync — Link transcription sessions to specific courses
• Document Editor — Export transcripts or lecture notes to a document
• Keyboard Shortcuts — Quick access to transcription controls