Scribd
Upload
18 views2 pages

Midpoint Project (BANA 6920) Kk237 Kyungrok Kim

The BANA 6920 Midpoint Project aims to address inefficiencies in meeting engagement by proposing an AI assistant that records dialogue, generates standardized notes, and tracks action items. The project will utilize Automatic Speech Recognition, Natural Language Processing, and Natural Language Generation to automate note-taking and enhance accuracy through Machine Learning. Team members will focus on developing specific models and datasets to support the AI's functionality, including public and synthetic datasets for training purposes.

Uploaded by

heisjinu
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
18 views2 pages

Midpoint Project (BANA 6920) Kk237 Kyungrok Kim

The BANA 6920 Midpoint Project aims to address inefficiencies in meeting engagement by proposing an AI assistant that records dialogue, generates standardized notes, and tracks action items. The project will utilize Automatic Speech Recognition, Natural Language Processing, and Natural Language Generation to automate note-taking and enhance accuracy through Machine Learning. Team members will focus on developing specific models and datasets to support the AI's functionality, including public and synthetic datasets for training purposes.

Uploaded by

heisjinu
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 2

Team Members: Alan Kim, Grey Kim, James Conahan, Joshua Andrada

BANA 6920 Midpoint Project

Business Problem:
During meetings in the professional workplace, when individuals focus on Meeting Minutes or taking notes,
this results in lack of focus and weakens engagement. To enhance efficiency and promote standardization, the
proposed solution would be an AI agent/assistant that is able to record dialogue, create standardized notes, and
provide follow-ups on deliverables mentioned in the meeting.
Potential Mock-Ups or Verbal Description:

1. Integrate Voice Recording Software through Python coding


2. AI analysis of transcripts derived from the voice recording
3. Create standardized notes & list of action items in an email-ready format
After initiating the AI agent through a pre-set / manual timer or certain key phrases, AI would automate voice
recording to text conversion & transcript analysis through utilization of Automatic Speech Recognition,
Natural Language Processing, and Natural Language Generation. Implementation of A.S.R. is necessary in
order to filter background noise but to also output a raw transcript that identifies and labels all the potential
speakers within the defined setting. N.L.P. will then be utilized for AI to perform analysis on raw transcripts
which allows for potential action items and provides context for standardized notes. N.L.G. then follows up in
order to create executive summaries and notes in an email-ready format. To enhance accuracy, Machine
Learning will be utilized to fine-tune the models that allow for ASR, NLP, and NLG implementation.

Description of the Datasets to Be Acquired or Generated


To support the development and training of our ASR, NLP, and NLG models, we will use both public and
synthetic datasets. Team members (Grey & Josh) will focus on two datasets each.
1. Public AMI Meeting Corpus:
A structured dataset of recorded meetings with transcripts and speaker labels. Ideal for training our ASR
and
NLP models to identify speech patterns and extract actionable insights
2. Synthetic Voice-to-Action-Dataset:
Create simulated meeting records with scripted action items and summaries. These will be annotated to
help
fine-tune our models in recognizing tasks and generating concise notes across varied dialogue styles
3. Sentiment and Intent Analysis Dataset:
Determines the emotional tone of text or speech. It can classify content as positive, negative, or neutral.
Techniques include using machine learning models and lexicons to analyze world choice.
4. Task-Oriented Dialogue Dataset:
Designed to train AI systems in structured conversations with a goal in mind. These datasets can allow
you
to develop appointment scheduling systems, customer support, and retrieval. An example of this dataset
would
be: MultiWoz (Multi-Domain Wizard of Oz).

Potential Task Division


(Alan): Develop ASR model (with voice capturing), test model for transcript creation & integrate background
filter, and visual design (support)
(Josh): Develop NLP model that allows for transcript analysis & derive insights and visual design (support)
Team Members: Alan Kim, Grey Kim, James Conahan, Joshua Andrada
BANA 6920 Midpoint Project

(Grey): Develop NLG model that integrates NLP model to create summaries & action items, and visual design
(support)
(James): Integrates the full flow (ASR → NLP → NLG), visual design (primary)

You might also like