DEX Audiobook

DEX existed as a Swedish book, but no English translation or English-language audio version existed.

Together with the author, I wanted to explore whether modern AI could be used to make the story available to an English-speaking audience — without losing the tone and character of the original.

The challenge was therefore bigger than just translating the text. The story had to work as literature in English, and then be turned into a coherent listening experience.

The Swedish edition was used as the source. Several AI services were used during the translation work to compare phrasings, linguistic rhythm and tone.

No single model was treated as the answer key. Instead, the models were used as alternative editorial voices. The author and I could compare their suggestions and judge which phrasing best preserved the original's:

The approach became a combination of machine translation and human editorial judgment. The AI produced alternatives; we were responsible for the linguistic and creative decisions.

Once the English text was finished, the audio version was created locally on a Mac Mini M4.

Processing locally made it possible to experiment with generation and post-processing without sending the entire production to an external cloud service. It also gave practical experience of what local AI can handle, where the quality limits are, and which parts still require human control.

The work involved more than generating a voice. A working audiobook requires consistency across a long story:

The result was published through purpose-built apps for both iOS and Android.

The project thus covered the entire chain:

The project showed that AI can make translation and audio production accessible to projects that might otherwise never have had the resources for an international edition.

At the same time, it became clear that AI does not eliminate the need for editorial judgment. Models can quickly produce many plausible phrasings, but they cannot decide for themselves which one best represents the author's intent.

In the same way, local audio generation can deliver a complete technical production — but that does not automatically mean the voice meets the emotional and sonic quality a commercial audiobook requires.

The current version shows that the whole production chain works. The next step is to evaluate whether a specialized cloud service can improve:

The existing production therefore serves both as a published product and as a baseline for the next level of quality.

DEX shows how I work with new technology: not by just evaluating a model or building a prototype, but by building a complete production chain and taking the result all the way to the user.

The project combined generative AI, human and machine language processing, local AI execution, audio production, product development, native apps for iOS and Android, and publication on the App Store and Google Play.