ChapterIT

Inspiration

People usually spend some time after a meeting to collect the key take aways. Integrating this app with any meetings platform such as Google Meet, Microsoft Teams, Zoho Cliq, etc… will avoid each one spending time on this. I take this as an inspiration and developed this AI Powered tool.

What it does

Automatic Chapters for Audio, Video & Meetings Platform

How we built it

It consists of 4 Modules

1. Audio Download from any Video or Audio Platforms
2. Transcript Generation - OpenAI/Whisper - https://github.com/openai/whisper
3. Topic Segmentation - Unsupervised Topic Segmentation of Meetings with BERT Embeddings - https://arxiv.org/pdf/2106.12978v1.pdf
4. Chapter Prediction - Using Custom Trained T5 Transformer Model - https://arxiv.org/pdf/1910.10683v3.pdf

Challenges we ran into

1. For Topic Segmentation I had two methods BERTopic & Unsupervised Method. For Time constraint, I moved with Implementing one
2. Pretrained Summary Models are not suitable for Chapter predict. Hence We trained a custom model to predict chapters.

Accomplishments that we're proud of

Finally I have built a AI powered Tool for Chapter Prediction with two sleepless nights & days.😇

What we learned

Learned How to maintain a Software project.

What's next for ChapterIT

• As part of MVP, I have currently added support for Youtube Platform. In future, will add support for other video & audio platforms.
• Add support for Meetings Platform

Built With

• deep-learning
• docker
• natural-language-processing
• python
• streamlit