Hi everyone,

Welcome to the 12th episode of "The Developer Digest".

The goal of this series is to achieve my goals, document my stuff, building things, experimenting with new stuff, learning and sharing every week.

Logs

Zerobase

Although, week 3 passed and today is week 4 first day.

But, i previously uploaded video on YT channel, talking about week 1 and 2 progress on how i picked the idea, build mvp and share it with beta testers to get initial feedback.

https://www.youtube.com/watch?v=egjusih8ftk

FYP Externals

So, my fyp externals just completed.

It was good, like most of time were spent waiting for evaluators while they were busy taking other evaluations.

We had 2 evaluators that take our evaluation.

First one was simple, cz the evaluator had launched his startups before and knows this problem actually exists. He further gave some ideas on how we can improve this product more, what else we can add to make it ready for market and launch.

Btw, i have my photo in poster in implementation screen ;)

and i also got a haircut ;)

Visual Speech Recognition

Yesterday, I tried a new project. So, long time ago, one of my friend gave an idea to train AI model, that takes slient video of person talking in camera, and model will predict what person is saying. Back then, we don't have any resouces mainly gpu to perform this experiment.

Previous week, i got a chance to experiment with AMD GPU, and I tried this Project.

tbh, it was great experiece.

I will share a bit here becuase i will share complete info about this in next video.

So, firstly i found dataset that i can use for this purpose. First step was to build a dataset pipeline so dataset can be downloaded and ready for training.

This includes downloading the speaker videos and their corresponding captions, converting videos into frames, extracting their lip part on each frame. Then I finetuned Resnet18 + LSTM on that dataset and overall got good results.

So, I got Character Error Rate (CER) of 0.12 and Word Error Rate (WER) of 0.24.

Weekly Picks

• https://www.notion.so/tta/How-to-Make-Your-Work-Visible-as-a-Founder-35b6a81416d1807e8701f917760fbe6a

• https://ramx.in/blog/the-internet-romanticized-exhaustion

• https://hasifff.substack.com/p/the-art-of-ending-things-before-it

• https://phoebedodds.substack.com/p/how-to-build-a-personal-brand

• https://www.youtube.com/watch?v=v6MWNrVbM4E

---

That’s it for this one.

Hope you like this episode.

See you in the next one ;)

- Psycho