meeting-quality-scorer
Score meeting quality from a transcript.
Quick Start
pip install -r requirements.txt
# Score a meeting transcript
python scripts/score_meeting.py --input meeting.txt
# Or with custom output paths
python scripts/score_meeting.py --input meeting.txt --out-md my-report.md --out-html my-report.html
Configuration
Set LLM backend via environment variables:
export MQS_BASE_URL=http://localhost:11434/v1 # Ollama
export MQS_API_KEY=ollama
export MQS_MODEL=qwen2.5:72b
Or create ~/.config/meeting-quality-scorer/config.yaml:
base_url: https://api.openai.com/v1
api_key: sk-...
model: gpt-4o-mini
Three Dimensions
| Dimension | Weight | How scored |
|---|---|---|
| 决议明确度 | 40% | LLM detects topics + decisions + owners |
| 时间效率 | 30% | LLM identifies off-topic / filler segments |
| 参与均衡度 | 30% | Gini coefficient on speaker word counts |
If no speaker labels → participation skipped, weights redistribute to 60/40.
Privacy
Transcript is sent only to your configured LLM endpoint. No telemetry.