2026-02-15
I love audiobooks. They give me access to books when I cannot or do not want to read in the conventional way. Sometimes my eyes are too tired. I often do things that require my eyes or hands but not much attention. If I am not in the mood to listen to music, or just sit and let my mind empty out, an audiobook is the perfect thing. Unfortunately, at time of writing, they only exist for very mainstream, very popular texts. You can seemingly find every ghost-written celebrity autobiography in audiobook form. Major history works on widely-studied events are often missing.
For the last several years, it has been possible to take a scanned PDF of a book and convert it into an audiobook, in an entirely automated way. The PDF can be run through optical character recognition (OCR) software to get an accurate plain text copy of the book's contents. This can be fed into voice generation software to get a complete audiobook. The quality of the narration is inferior to a professional voice actor and it probably will remain this way for years. It is comparable to an amateur human narrator. This is impressive though, good enough for me and many others to enjoy these audiobooks.
People are already producing machine-voiced audiobooks, but they are doing the steps above by hand, so their output is limited. They post their work online in various places. I appreciate the work these people put in, but the process should not require work by human beings. From a purely technical perspective, we should have a program that takes a PDF, lets you choose a voice, then spits out an audio file.
If you are in a position to make this a reality, maybe you want to go for it. Otherwise, one of these days, I will have to do it myself.