Firn for audiobooks
Hear your book tonight.
You have the chapters. What you don't have is a booth, a narrator's calendar, or the budget a full studio production wants before a single listener presses play.
Firn narrates your drafted chapters straight from the same world bible your book reads from (narration in one voice, each character in their own), so you can listen to the whole thing the day you finish writing it.
Said plainly: this is machine narration: modern neural TTS, not a human voice actor. It is a fast, readable draft-listen and a workable budget narration, and we would rather tell you that than dress it up.
Free to start. Listen inside the studio for free; the downloadable listen package is a premium export.
Listen in the studio
Press Listen. Hear it. No export required.
Once a chapter has drafted prose, Read mode unlocks. Its Experience Listen panel plays your chapters aloud with chunked playback (narration and character voices) right where you are reading. You hear the pacing, the dialogue, the places a sentence trips, before you commit to anything.
It is the fastest way to proof a draft with your ears instead of your eyes: the read-through you would otherwise pay for, available the moment the words exist.
A cast, not a monotone
Narration in one voice. Each character in their own.
Firn segments each chapter into narration and dialogue, then assigns voices from a curated catalog: your narrator gets one voice, and each speaking character is cast to their own. The result reads as scenes with people in them, not a single flat read of the whole page.
Narration and dialogue, split
Every chapter is broken into speakable segments (what the narrator says, and what each character says), so the voices land on the right lines.
One voice per character
You cast each character from a set of voices, and they keep that voice all the way through — the same person sounds like the same person in chapter one and chapter thirty.
Every line, and a running order
A listen package you own: the audio, plus the running order to it.
You get a folder of audio, split up line by line, plus a plain list naming every piece: which chapter it came from, who read it, and whether it is narration or someone speaking.
- A running order of every clip, with the voice and the speaker named
- One file per line, so you can re-record or re-order a single moment without redoing the whole book
- Narration and spoken lines marked separately, so a human narrator coming in later knows exactly what to re-record
- It is source material, not a locked file: yours to process further, master, or replace segment by segment
Nothing unusual about any of it. The audio is just audio, and the running order is a plain text file, so whatever you already edit with will open it.
{
"title": "The Glass Orchard",
"voice_catalog": "kokoro",
"chapters": [
{
"id": "ch2",
"title": "The Greenhouse",
"segments": [
{
"file": "ch2_0001.wav",
"role": "narration",
"voice": "narrator_warm",
"seconds": 14.8
},
{
"file": "ch2_0002.wav",
"role": "dialogue",
"speaker": "Elara",
"voice": "af_bright",
"seconds": 6.2
},
{
"file": "ch2_0003.wav",
"role": "dialogue",
"speaker": "Mara",
"voice": "am_low",
"seconds": 5.1
}
]
}
]
}A job you can watch
Kick it off. Watch it run. Come back to a finished narration.
Narration runs as an async job with live status (pending, running, completed), so you are not staring at a frozen tab. Segments synthesize concurrently, and a segment that fails is retried rather than quietly dropped, so the finished job is the whole book, not most of it.
Kokoro neural TTS, through OpenRouter
Synthesis runs on Kokoro, a modern neural text-to-speech model. It is machine narration: clear and steady, positioned as a draft-listen and a budget narration, not a stand-in for a human read.
Concurrent, with retry
Several lines are recorded at once so you are not waiting all afternoon, and if one fails it is simply tried again — you do not end up with a silent gap in the middle of a chapter.
The same world, read aloud
The audiobook reads the same canon as every other door.
There is one world bible. The audiobook narrates the very chapters your book exports, the same cast your visual novel and comic draw from. Rename a character in the Codex and it is that name the narrator reads, because the story the voices tell is the one story Firn holds, not a separate copy that drifts.
The prose those voices are reading is your book. Adapting for another shape? See all formats.
Call it what it is: AI narration.
Listeners can tell, and the ones who care about it care a lot. Firn does not pretend otherwise. This is neural TTS (a fast way to hear your draft, and a real budget option for a narration you can ship), and it is not a human voice actor.
Use it to hear your own prose and catch what your eye skips over, to offer a listenable version without a studio budget, or as the guide track a human narrator later re-records line by line. And when a platform asks whether the narration is synthetic, you can answer precisely, because the running order already says who read what.
You are the author. The crew narrates on your say-so; you choose the voices, approve the run, and own the result. Assistant, not vending machine.
Same world, other shapes
One bible. A shelf of formats.
The audiobook is one way out of your world. The chapters those voices read are your book; the same canon exports as a visual novel, a comic, a webtoon. Every door opens from the same core, no rebuilding the cast.