LLM Audiobook Prompts

I do most of my reading while commuting in Taipei, and the course textbook has no commercial audiobook. So I sat down with an AI assistant and built one. The whole method is below — every prompt, the script, and the reasoning — because the useful part isn't my audio files, it's that you can do this with any book you own, in about an hour.

The method, in eight steps

1. Buy the textbook

   Non-negotiable, and it's step one for a reason. Everything below assumes the book is yours.

2. Obtain a PDF version

   Many publishers sell a PDF or ePub directly. If you bought physical, check whether your purchase includes digital access, or whether your library licenses an accessible copy.

3. Point the assistant at the file, and ask the real question

   Don't ask “make me an audiobook.” Ask it to evaluate the options first. That one framing change is what produced an honest answer instead of a confident wrong one.

4. Let it discover its own limits

   My assistant tried to install a cloud text-to-speech tool and found its sandbox had no network access to speech services. That failure was the most useful step in the project — it moved the work onto my Mac, which is where it belonged anyway.

5. Reject the tempting wrong tool

   NotebookLM was my first instinct because it's free and makes great audio. But it produces a conversation about your document — two hosts chatting, paraphrasing. For studying a textbook I needed the actual sentences, in order. Faithfulness beat production value, so we used the plain say engine built into macOS.

6. Extract chapters to clean text

   The assistant read the PDF, found each chapter's real boundaries, stripped page numbers and running headers, and wrote one Chapter_01.txt per chapter into a folder. Chapter 1 came out at 8,368 words — about 56 minutes of audio.

7. Narrate with one double-click

   A tiny shell script walks the folder and converts every chapter to an .m4a. No terminal commands, no account, no API key. Free and offline.

8. Pick a voice you can stand for 12 hours

   The default (Samantha) put me to sleep in four minutes. Swapping to Zoe (Premium) made it listenable. Download voices at System Settings → Accessibility → Spoken Content → System Voice → Manage Voices, then change one line in the script.

The prompt — copy this

Paste this into any AI assistant that can read files and run code on your machine. Fill in the two bracketed bits.

I own a copy of [BOOK TITLE + AUTHOR] and I have it as a PDF at [FILE PATH]. There is no
commercial audiobook and I want to listen to it while I commute. It is for my own study use
only; I will not distribute the audio.

Before you build anything, evaluate the options and tell me the trade-offs:
1. Can you generate speech yourself, or is your environment network-restricted? Test it and
   tell me what actually failed rather than assuming.
2. Compare: (a) NotebookLM-style tools, (b) a paid cloud TTS API like ElevenLabs or OpenAI,
   (c) the free offline text-to-speech engine built into my operating system.
   I care most about FAITHFULNESS -- I want the book's actual sentences read verbatim, not
   summarized or turned into a discussion between hosts.
3. Tell me whether the length of the book creates any token or context limits, and why.

Then, once I have picked an approach:
- Extract each chapter from the PDF into its own clean .txt file. Find the real chapter
  boundaries, and strip page numbers, running headers, and footnote noise so the narration
  does not read junk aloud.
- Write me ONE double-clickable script that converts every chapter file into an audio file in
  the same folder. I do not want to type terminal commands. Include a clearly commented line
  at the top where I can change the voice and the reading speed.
- Do chapter 1 first as a sample so I can check the quality before you process the whole book.

Tell me the word count and estimated audio length for the sample chapter.

The script

This is the entire thing. Save it as Narrate_Textbook.command next to your Chapter_*.txt files, then run chmod +x Narrate_Textbook.command once (or ask your assistant to). Double-click it. First run, macOS may warn about an unidentified developer — right-click → Open.

#!/bin/bash
# Double-click this file to narrate each Chapter_*.txt into an .m4a audio file.
# Free, offline, uses the macOS built-in `say` voice. No account needed.
cd "$(dirname "$0")"

VOICE="Zoe (Premium)"   # Change me. Download voices: System Settings > Accessibility >
                        # Spoken Content > System Voice > Manage Voices.
                        # If say errors on the name, run:  say -v '?'  to list installed voices.
RATE=175                # words per minute; lower = slower

shopt -s nullglob
echo "Starting narration with voice: $VOICE"
for f in Chapter_*.txt; do
  base="${f%.txt}"
  echo "Narrating $f  ->  $base.m4a   (a few minutes per chapter)"
  say -v "$VOICE" -r "$RATE" -f "$f" -o "$base.m4a" --file-format=m4af --data-format=aac \
    || say -r "$RATE" -f "$f" -o "$base.aiff"
done
echo ""
echo "Done. The .m4a files are in this folder. You can close this window."

The actual conversation

Here is the real exchange, lightly edited.

Hide annotations

Transcript lightly trimmed for length and to remove an unrelated request about coursework. Tool-call details are collapsed to Ran a command markers.

Friend, M. (2021). Interactions: Collaboration skills for school professionals (9th ed.). Pearson. Narrated with the macOS say engine (voice: Zoe Premium) for personal study use; audio not distributed.