audio player screen
library page
audio versions
web UI in pdf parsing step
TTS audio generation pipeline

Lots of young people nowaday are more addicted to social media content, those content are easy to find, fun to see and abundant. But somehow we forget the joy and the meaning of reading books, and we miss our own goals of reading more books year after year, further more, those social media content aren't really benefit us, some content we see once and forget, but we lost minutes or hours for them.

I am also in that situation, I used to passive scroll feeds and almost forgot to read books, even when they are in my bookshelf and easy to reach. As a person who love improving himself, I can not just stay in the comfort zone and keep on enjoying the social media content that mostly don't bring me any lasting values, only fatigue and sleep late.

So I was thinking of making audio book, to help me and other people to skip scrolling passively, relax our eyes by not having to look at the screen for long time, use ears instead, all I have to do is to listen to audio books, and my hand can do other things like chores or cooking.

What it does

1) The backend - the heart of the project:

A pdf parsing, audio generation pipeline that helps me to parse every pages of pdf (epub and docx sooming soon) documents. When document content are done with parsing and verified by human, admin can designate an AI voice for audio generation, then the pipeline will take over and start producing audio content from the first phase.
- Backend also serves as a online library management so admin can manage books, authors, genres, pricing, publishing state of every books.

2) The mobile app

End user application product to help end users reach the generated audio books from backend.
The app has in app purchase feature (with support of revenuecat SDK) which charge users for monthly or yearly subscription.
The app helps users to easily manage books, create favourite collections, searching for the books they want to listen, and submit requests to admin for production of his favorite books that don't exist on the store.
The app supports offline listening, so subscribed users can pre-download ebook locally to the phone, then listen to it later when he does not have internet connection.
The mobile app gives user 14 day free trial use to listen to any book for free, it only requires small amount monthly subscription and will never shows in app advertisement to users.

How we built it

The initial idea emerged since 2025 when author got to knew there are tts tools that can help produce audio from text. After a few months of research, try and fail, author finally can build and finish this product nicely and working.

Challenges we ran into

1) Original tts tool can only produce short audio content by copy/paste processed and normalized text, the action is not fun and repeatedly, normal people will give up just by having to copy paste every paragraph of an pdf book into that tool. And human labour under a boring situation is fastest way to error prone, low quality audio product.

2) Lots of pdf extract tools can fully get all text from a pdf book, and we can feed this text into the tts pipeline. But what if during listening users find a mis-pronounciation, un-wanted content that should not be brought into audio, and lots of unwanted things like footnote, superscript, subscript, watermark text and lots of text is not audio generation friendly like date of birth, money amount, scientific notation, or some paragraph ending with a dash to mark the concatenation with the next paragraph. All those things are obstacles that can together making end result unacceptable.

3) Some problems like text punctuation chunking or prosody correcting are other problems that can contribute to a nice audio content

Accomplishments that we're proud of

1) Working web backend with the working pipeline that can produce audio content from pdf ebook properly. It is still not perfect but the fraction is very small, compared to direct, raw processing. 2) A friendly flutter mobile app that is simple to use and completely can work when screen turned off, does not eat much battery, ram usage. A beautiful listening streak feature that encourage users to keep passively listening to unlock more badges.

What we learned

1) Writing code with the help of AI is easy, but making a good and sustanable product is stll very hard. 2) Building a tool as a way to make for a living is very hard, and you have to do lots of things, both technical and non-technical in order to be able to make the app go live and working smooth. 3) And from the moment the app go live, money won't easily flow in, and the maintainance, customer support, product refining treadmill will start working forever.

What's next for Audiu

1) If this project fail, it is still a tremendous project author have ever built and I will continue on making it perfect for me and future customers. There will be more and more good features to add. And the ultimate result will be I can read more and learn more from more books on the way.