We believe that …
We can use a deep learning process called automatic speech recognition (ASR) to convert speech to text quickly and accurately in order to generate meeting transcriptions, which can then be quickly and easily converted into concise and accurate meeting notes.
How critical is this hypothesis?
We aim to use this research to …
Discover if it would be time-and-cost effective to use a speech-to-text service, such as Amazon Transcribe or Microsoft Team’s built-in transcription facility, to assist in writing meeting notes.
To verify that, we will…
- Hold an online meeting with the speech-to-text service “listening” and generating a transcript. This will of course be with all participants’ knowledge and agreement.
- During this meeting a nominated person will act as a scribe to manually take notes. They will keep these notes private. The notes are the control in this experiment.
- Get the scribe to record how much time it took them to manually write the notes.
- Use Speaker Diarisation to identify who said what from the transcript.
- Use a tool / interface to highlight sections of the transcript and to generate a summary of the key decisions and actions.
- Record the time taken to create the meeting summary from the transcript.
- Get the meeting attendees to score each set of notes for clarity and accuracy without knowing how they were generated.
- Repeat this process until a statistically significant number of example meetings have been held and recorded.
- Compare the quality of machine-assisted notes with manual notes, taking into account the time and cost of each method.
- Time spent on each method
- Cost of each method
- Quality of each method (based on accuracy and clarity scores)
We are right if…
After taking into account labour costs, machine-assisted meeting notes are significantly cheaper to generate than manual ones without sacrificing quality.
Non-urgent advice: Experiment #004
Text to speech for meeting notes.
- (Not yet complete)
Based on the Strategizer lean test card