# How I built an AI podcast generator that turns PDFs and articles into audio conversations

> Published 2026-11-17 · https://www.promptzone.com/gaspard_bak/how-i-built-an-ai-podcast-generator-that-turns-pdfs-and-articles-into-audio-conversations-5i1

Most of what I have to read arrives as a PDF, a long article, or a page of course notes. I built Podcastify because I kept saving that material for a free hour at a desk and never finding one. The product turns a document into a two-host conversation I can listen to on a walk, and it has been running since 2024.

I am Pierre Kasparian, a software engineer and AI engineer. Podcastify started as a personal study tool while I was studying computer science engineering at Université de technologie de Troyes and working as a freelance AI engineer. The first version read my own notes aloud. It was hard to follow, which sent me looking for a better format.

### A conversation is easier to follow than a wall of text

Text-to-speech reads what is on the page. That works for a paragraph and falls apart over a report. One voice keeps the same pace and tone for twenty minutes, and the listener gets no help separating the main argument from the examples.

Podcastify asks a language model to write a dialogue first. One host introduces the topic, and the other asks the questions a listener would ask. That structure does the work a reader normally does: it marks what matters, restates a point in different words, and moves on when the topic is covered. The result sounds like two people discussing the source, not a machine reading it.

### Two phases, and a script you can edit

The pipeline has two phases. A multimodal language model reads the source and writes the transcript between a Host and a Guest. A text-to-speech engine then renders each turn with a different voice and joins the parts into one MP3. Most episodes finish in under two minutes: about 30 to 60 seconds for the script, then another 60 to 90 seconds for the audio, depending on length.

The order matters. The transcript appears before the audio is generated, so the user can fix a weak opening, remove a repeated point, or change the tone before spending audio characters. Editing a script is cheaper than regenerating audio, and that control is the main difference between a podcast generator and a button that reads text aloud.

Inputs are text, PDFs, public webpage URLs, images, and YouTube links. The output is one mixed episode as an MP3, an editable transcript, a public share link, and an iframe embed code. Intro and outro music can be added in one click, with a 300 ms fade at each join. Episodes can run up to 25 minutes.

### Voices, languages, and the stack

Podcastify includes more than 20 premium voices. A user can clone a voice from a 10 to 90 second sample, with 30 to 60 seconds giving the best result, and hold up to two cloned voices per account. Cloning helps creators who want their own voice on every episode and teams that want one brand voice across a series.

The interface and the generated audio support 15 languages, so one source can become several versions without re-recording or hiring voice actors. Commercial use is allowed, and the user keeps ownership of the generated audio.

The stack is Next.js on the front end, with FastAPI, Celery, and Redis on the back end. Scripts run on Google Gemini, and the voices come from several TTS providers.

### Who it is for

The product fits people who already publish written work and want an audio version: marketers turning blog posts into episodes, teachers turning notes into lessons, newsletter writers offering a listenable edition, and teams summarizing reports. It also fits anyone who would rather listen than read, which is where it started.

The Hobby plan costs $9.95 per month and includes 170,000 audio characters per month, every input type, the full voice library, and two cloned voices. Billed annually, the same plan is $8.59 per month, or $103.08 per year. A $6 top-up adds 200,000 characters that never expire, and a one-time trial pack lets a new user test the full workflow with three transcripts and one audio.

Generation runs on the paid plan. There is no free tier, though several free browser tools work without an account, including Transcript to Podcast, a podcast script generator, and a course summarizer.

### About Podcastify

Podcastify is an AI podcast generator that turns text, PDFs, webpages, images, and YouTube videos into two-host episodes. You can try it at [Podcastify](https://podcastify.io).