Back to list

Backlist

Paste in a finished chapter; two open models in your browser split the job: one reads aloud, one listens back and flags mismatched lines.

汽水开盖日记
Author
@汽水开盖日记
Project ShowcaseContent & MediaOnlineclaude-code
4
0

Project intro

Backlist is an audiobook tool for independent authors: with a dozen books and no budget for a human narrator (200 to 400 dollars an hour), it lets an open speech model in the browser read a chapter aloud first, then a second model listens back and checks it line by line against the text.

The proof-listening step is the point: you don't listen from start to finish, only the mismatched lines are flagged — for example Chapter IV read as “Chapter Roman Four”, or a long sentence cut off halfway. If a reading sounds right you approve it; if not, you fix the text or add a name to the pronunciation guide, and only that line is re-read, not the whole chapter.

At the end you can export chapter audio processed to audiobook spec: pauses trimmed, loudness leveled; there is no server, and your manuscript never leaves the browser tab. First use requires downloading the models, about 400MB on a GPU machine, about 170MB on a CPU-only machine.

Two honest caveats: the author is upfront that Audible's ACX currently requires human narration, so this output fits Google Play, Kobo, or direct sale better; and there is a limit that can't be worked around — the checking model tidies away repeated stutters, so a few repeated-word errors can slip through, which the author states in the documentation.

Actions

Visit project URL

Tech stack

transformersjs

Project info

Project category
Content & Media
Project status
Online
Collaboration intent
None

Comments (0)

ME
0/1000
Loading comments...