Here’s the problem I started with: my blog content is long, and realistically, not everyone has time to read it. So in this video I tested how to make a website “talk” by generating an audio version of each article using AI voice cloning. I broke it into three parts: cloning my voice with a text-to-speech model (Fish Audio), generating a portfolio/blog site I actually control (so I can add my own APIs), and then deploying the whole thing to a VPS so I’m not boxed in by typical serverless limits. I cloned my voice in seconds, then upgraded to a longer training sample (a few minutes) and the result was honestly kind of scary-good. From there I grabbed an API key, tested the endpoints in Postman inside Cursor, and learned the key detail: you need to pass the reference/model ID to get your actual cloned voice. Then I “vibecoded” the integration by getting Cursor to build a component that takes blog text, calls Fish Audio, and returns an audio file right on the page. Finally, I hosted the Next.js app on a Hostinger VPS using a simple build + `next start` flow. It’s not fully production-hardened (you’d still want HTTPS, a domain, port 80, etc.), but it shows how you can get full control over deployment and costs while adding AI voice to real web projects.