Turn Anything Into An Audiobook With Your Own Voice

My mom was never going to read 50 pages of my guides.

She’d been asking for months what I actually do all day, but reading a stack of guides isn’t how she takes in information.

So before a road trip, I cloned my own voice and turned my AI guides into a single audiobook, narrated by me. She listened to the whole thing in the car and texted me that she finally understood what I do. She assumed I spent five hours recording it. It took three minutes.

Now I turn long business research, AI newsletters and every article I save and never open into audio I listen to while I walk or drive. Here’s the exact setup, the prompts I use, the real pricing and the limits.

The Deadline

The free part is Fish Audio’s S2.1 Pro Free API model, and free access now runs through November 30, 2026 (it’s been extended three times). After that you’d use the paid S2.1 Pro model at $15 per million bytes of text, which Fish says is about 180,000 English words, so a long book costs a few dollars. (Fish Audio)

No credit card. Fish says the free model is free to call with no card required.
No hard character cap. It runs under a Fair Use Policy, so a book-length document is fine.
The same API as the paid version. You switch by changing one header from s2.1-pro-free to s2.1-pro.
83 languages, so you can narrate in a language you don’t speak.
Your own cloned voice works with it. You pass your voice’s ID with each request.

The website plans are separate from all of this. You don’t need a subscription to use the API. On the consumer plans, Free is $0 (about 7 minutes of audio a month, personal use only), Plus is $11 a month billed yearly ($15 monthly), and Pro is $75 a month billed yearly ($100 monthly).

Know This Before You Upload

Fish says free-tier requests may be used to improve its models, there’s no uptime or speed guarantee, and products making over $1M in annual recurring revenue should contact them first. Fine for your own listening. Don’t send it anything confidential or put a paying customer on it.

You don’t need to know how to code. You need an account, a key and a chat with Claude. If you have Claude Code, it does the last step for you.

Get your account and API key

Make an account at fish.audio, then open API Keys in the developer dashboard and create a key. It’s only shown once, so copy it right away and treat it like a password.

Clone your voice or pick one

Clone your voice. Fish recommends 2 or 3 clips of 15 to 20 seconds each, recorded in a quiet room with small pauses between sentences (their tips). Read normally. If you read in a big announcer voice, everything it makes sounds like that. Set the voice to private, and copy its ID from the voice’s page.

Or use a library voice. Fish has a big community library, including documentary-style narrators. I cloned my voice for my mom because the point was that it was me. For my own newsletters I’d rather have a narrator. Hours of my own voice would be strange.

Have Claude rewrite it for the ear

This step decides whether you get an audiobook or a robot reading a spreadsheet. Raw documents are full of headings, bullets, links and tables, so you hear "bullet, bullet, h t t p s." Upload your document to Claude and paste this in.

Document To Narration Script

I uploaded a document. Rewrite it as a script that will be read out loud by a text to speech model. This is going to become an audiobook, so it has to work for the ear, not the eye. Rules, follow all of them: 1. Output plain narration text only. No headings, no titles, no section numbers, no markdown, no asterisks, no bullet symbols, no dashes used as list markers, no tables. 2. Where the original used a heading, replace it with a short spoken transition instead. Something like "Next, the part about pricing." Never read a heading as a label. 3. Where the original used a list, turn it into a spoken sentence. "There are three things that matter here. First... Second... Third..." Never read symbols aloud. 4. Remove every URL, link, footnote marker, and citation number. If a link mattered, say the name of the source out loud instead, like "according to their pricing page." Never spell out a web address. 5. Spell out anything that would be read wrong. Write numbers, dates, currency, and units the way a person would say them. Write "twenty twenty six" not "2026". Write "fifteen dollars a month" not "$15/mo". Expand abbreviations the first time, so "annual recurring revenue" instead of "ARR". 6. Break long sentences into short ones. If a sentence would make a person run out of breath, split it. 7. Keep all the real information. Do not summarize, do not shorten the substance, do not add anything that was not in the document. Only change the form. 8. Use paragraph breaks where a narrator would naturally pause. 9. Do not add an intro, an outro, or a sign off unless I ask for one. Start with one short spoken opening line that tells me what I am about to listen to. Then give me the full narration text and nothing else. No preamble, no notes, no explanation of what you did.

Read the first few lines of what comes back. If it still has a heading or a stray symbol, say "line three still has a bullet, fix it."

In Claude Code, or any AI tool that can run code, this is one more prompt. Set your key as an environment variable called FISH_API_KEY first, then paste this with your voice ID and a file name filled in.

Send It To Fish Audio

I want to turn the narration text above into one audio file using the Fish Audio API. Write and run a small script that does this for me. Details: - Endpoint: POST https://api.fish.audio/v1/tts - Auth: Bearer token in the Authorization header. My key is in the environment variable FISH_API_KEY. Read it from there, never print it, and never write it into the script. - Model: send the HTTP header "model" with the value s2.1-pro-free (free through November 30, 2026). If I tell you to use the paid model, use s2.1-pro instead. - Body: JSON with "text" (the chunk), "reference_id" set to [MY VOICE ID] and "format" set to "mp3". Requirements: 1. Split my text into chunks at paragraph breaks, about 2,000 characters or less each. Never split in the middle of a sentence. 2. Send the chunks in order, one at a time. If a request fails or returns 429, wait and retry with a growing delay (2, 4, then 8 seconds) before giving up. The free tier has no uptime guarantee, so occasional failures are expected. 3. Join the audio into one mp3 in the same order. Use ffmpeg if it's installed. If not, tell me and join the files another way. 4. Save the finished file as [FILE NAME].mp3 in this folder and delete the temporary chunk files. 5. Tell me the final file name, how long the audio is, and how many characters you sent. Optional: if I give you an audio file of my voice instead of a voice ID, first create a voice model with POST https://api.fish.audio/model (multipart form: title, voices = my file, visibility = private, train_mode = fast), wait until it's ready, and use the id it returns as the reference_id. Before you run anything, show me the plan in three lines so I can confirm.

Drop the finished mp3 into your phone or any podcast app that lets you add your own files.

No Claude Code?

Use Fish Audio’s website to generate the audio from Claude’s narration text. The free website plan caps each generation at 500 characters, so you’ll paste in short sections and stitch them yourself. The narration prompt is still what makes it sound good.

Everything you told yourself you’d read later.

Long research. Market reports and competitor breakdowns I’d never sit and read.
AI newsletters and saved articles. Your read-later folder is really a never folder. This empties it.
Things for other people. An onboarding doc, a family history, a set of instructions.
Study material and meeting prep. Review on the drive.

What made it stick is batching. Once a week, everything I didn’t get to becomes one long file.

Weekly Listening Session

I'm going to paste several things I saved and never got to: newsletters, articles and research. Turn them into ONE narration script for a single listening session. The pieces: [paste each one, with a line like "PIECE 2: [title]" between them] Steps: 1. Put them in the order that makes the most sense to listen to, grouping similar topics. 2. Before each piece, add one spoken line that tells me what's next and where it's from, like "Next, from the Monday newsletter, a piece about pricing." 3. Cut repeats. If two pieces say the same thing, keep the better version and say so in one line. 4. Follow all the read-aloud rules: no headings, symbols, links or tables; numbers and abbreviations spelled out the way a person says them; short sentences; paragraph breaks where a narrator would pause. 5. Keep the real information. Don't add facts or opinions that weren't in the pieces. Start with one opening line telling me how many pieces are in this session and roughly how long it'll take to listen (about 150 words per minute). Then give me the full script and nothing else.

The free window ends November 30, 2026. Set it up now and decide later whether the paid model is worth it.
Free means best effort. Requests can be slow or fail, which is why the second prompt retries.
Only clone a voice you have permission to use. Fish’s own guidelines say the same. My rule is to clone my own voice, tell people when it’s a clone, and never use it anywhere someone might act on it thinking it was really me.
It’s a very good machine voice, not a performance. For information it’s more than good enough. For something that lives or dies on delivery, record it yourself.

The Real Win

I get through roughly twice as much information now without adding an hour to my week. Pick the longest document in your saved folder, run it through the prompts above, and listen to it tomorrow.

Turn AnythingInto An AudiobookWith Your Own Voice

Turn Anything
Into An Audiobook
With Your Own Voice