Rebuilding the Audio Bible for v3

The Anselm Project Bible has had audio since v2. For v3 I redid it, and the part that took the actual work wasn't the narration. Text-to-speech handles that fine now. The work was getting the names right.

Getting the Names Right

Text-to-speech models do not know how to say most biblical names. The Bible is denser with transliterated Hebrew and Greek proper nouns than almost anything else you could hand a narrator, and most of them do not follow English phonetics at all. Feed a model Zedekiah, or Amittai, or Mahershalalhashbaz, and it guesses. The guess is usually wrong.

And biblical names are hard for people, not just for models. One of the most common things someone in my congregation asks me before they read Scripture aloud in worship is how to say the names in their passage. The names are difficult, and nobody wants to stumble over a prophet in front of the church. A narrator that says the names right means you can hear them before you have to say them.

The way I handled this before was to run each passage through AI first: one model picked out the hard names, another respelled them for phonetics, and the corrected text went to the narrator. It worked, but it cost time and tokens, and it was solving the same problem over and over for names that never change. For v3 I skipped that and built a name register instead. A name gets respelled once, I confirm it sounds right, and it is set — around 2,200 of them, fixed in a table instead of re-derived by AI on every pass. It is the least interesting thing I did and the thing that makes the rest worth listening to.

Hearing the Original Languages

The audio also reads the original languages, not just the English. Every verse has Hebrew in the Old Testament and Greek in the New. I have made the case in other posts that the original languages carry weight the translation can't, and I was not going to build an audio Bible that only let you hear the English. If you want to hear a passage in the language it was written in, it is there.

Inside the Reader

The other change in v3 is that the audio plays inside the reader instead of off on a page of its own. You press play on whatever passage you are reading, and the text follows the voice as it is spoken.

In English it follows word by word, because I can sync the highlight to the actual speech timing. In Hebrew and Greek it follows verse by verse. Word-level timing in the original languages is harder to get right, and pretending I had it would just produce a highlight that drifts off the word being spoken, which is worse than not having one. Verse by verse is honest, and for the original languages it is the part that actually helps — seeing the line while you hear it is how you start to recognize a script you do not read fluently.

A single toggle switches between English and the original, and your choice carries from one passage to the next. When you are listening in the original language the reader shows both, the original sliding in under each English verse so you can follow the two together. It plays through a passage and rolls into the next one on its own, so you can leave it running, and it answers to the spacebar and your phone's media controls like anything else you would listen to.

That is the update. It is live now across all sixty-six books. Open any passage in the Anselm Project Bible and press play, and if you want to see the reader it lives in, the reports in the Share Gallery come from the same place.

God bless, everyone.