Retake

Inspiration

I take part in a lot of hackathons, and every hackathon needs a demo video. For me, that's the hardest part. Recording my own voice always felt too hard, so I made a mistake: I let AI do everything for my videos. Then I realised a demo video can look a little less polished, but if a human records it, it feels more connected.

When I started recording myself, I had to practise my script again and again. And I never knew if what I was saying was clear to other people. I always just assumed it was, and submitted.

What it does

Here I can paste my script and practise it properly. Retake shows me which words I said wrong, and which words people might mishear. It also shows my fillers like "umm", long pauses in the middle of a sentence, and if I'm speaking too fast. Then I can retake only the lines that need it, and compare take 1 and take 2.

How I built it

How it works is simple. A speech to text service called Deepgram is the listener. It tells me every word it heard, when I said it, and how sure it was. Then my own code lines up my script with what it heard, word by word, and marks the differences. It's built with Next.js and it's live on Vercel.

I built it with the Build With AI skill pack. First I wrote the scope, then the PRD, then the spec. Then I built it in small steps, and I tested every step before the next one.

Challenges I ran into

My first real take came back almost empty. Every word was marked as not heard, even though I read the full script clearly. My mic wasn't picking me up. So I added a mic level bar, and a way to hear my own recording.

Then some words were marked red even though they sounded clear to me when I listened again. So now you can click any red word and hear exactly what the listener heard.

What I learned

I always just assumed people understood me. Now I can check it before I record. And this whole thing got better by using it myself.

Built With

• deepgram
• next.js
• node.js
• react
• tailwindcss
• typescript
• vercel
• web-audio-api