Meeting Transcription
Automated speech recognition and structured minutes generation with decision extraction.
Challenge
A major financial institution held dozens of meetings daily, with minutes recorded manually. Assistants could not capture all decisions and action items, and retrospective information recovery from recordings took hours. Deadlines, assignees, and decision context were frequently lost.
Solution
The system recognizes participants' speech, converts it to text with speaker identification (diarization). An NLP module extracts key decisions, action items, deadlines, and assignees. Structured minutes are generated with timestamps linked to the recording. Minutes are available within 5 minutes after the meeting ends.
Results
Technologies
Approach
Meeting format analysis
Studying meeting types, minutes requirements, and room acoustic conditions.
ASR setup with terminology adaptation
Training the speech recognition model on specialized financial sector vocabulary.
NLP decision extraction module development
Building a model that extracts key decisions, action items, deadlines, and assignees from text.
Pilot on real meetings
Testing the system in live conditions, collecting feedback, calibrating accuracy.
Similar challenge?
Tell us about your project — we will propose the optimal solution.
Discuss a project