Recording Flow
Understand the complete flow from recording to actionable insights.
Overview
EdgeNote AI processes your audio through a streamlined pipeline that transforms speech into structured, actionable content.
Recording
Capture Audio
- Mic + System Audio
- Meeting Detection
- Audio Levels
- Timer Display
Transcription
Whisper AI
- Speech-to-Text
- Speaker Diarization
- Timestamps
- Language Detection
Summarization
LLM Analysis
- Key Points
- Action Items
- Decisions
- Goals
Outputs
Results
- Summary
- Transcript
- Insights
- Export
Step 1: Recording
The journey begins when you start a recording. EdgeNote AI captures audio from your selected sources and displays real-time feedback.
- Microphone input (your voice)
- System audio (meeting participants)
- Or import existing audio files
- Live audio level visualization
- Recording timer
- Pause/resume capability
Step 2: Transcription
When recording ends (or you import audio), EdgeNote AI runs the Whisper model to convert speech to text.
What Happens:
Audio Processing
Audio is processed by the Whisper model using GPU acceleration when available.
Language Detection
Large models automatically detect the spoken language (99 languages supported).
Speaker Diarization
When enabled, the system identifies different speakers and labels their segments.
Timestamped Segments
Output includes precise timestamps for each segment of speech.
Step 3: Summarization
The transcript is then analyzed by a local LLM to generate structured summaries and extract insights.
The LLM reads the full transcript and generates a concise summary following your selected template format.
Key points, action items, decisions, and goals are automatically identified and categorized.
Step 4: Outputs
The final results are saved and made available for review, search, and export.
Summary
Structured overview
Transcript
Full text with timestamps
Export
Text, Markdown, PDF
Processing Time
Processing speed depends on your hardware and model choices:
| Hardware | 1 Hour Recording | Notes |
|---|---|---|
| Apple Silicon (M1-M4) | 2-5 minutes | Full GPU acceleration |
| NVIDIA RTX GPU | 2-5 minutes | CUDA acceleration |
| AMD/Intel GPU | 4-10 minutes | Vulkan acceleration |
| CPU Only | 20+ minutes | Use smaller models |