The slowest part is uploading each section to LingQ, which took me a few hours

Open Forum
4

I did toy with this idea for a couple evenings and I found a way to solve some cases. I will copy some of the docstring here so you can decide if this fits your particular case.

You may find yourself in one of this cases:

(1) Your texts and audios are chapter-structured.
    Then you don't need any of this and you can just post your pairing to LingQ.
(2) Your texts are chapter-structured but your audios are NOT.
    In this case you would want to split with "Preserving chapters" with simply:
    process_files(
        language="ell",
        files_dir="my_input_files",
    )
    This only modifies audios.
(3) Your texts and audios are NOT chapter-structured.
    If you want them to be chapter-structured, this (and most likely no tool out there) does
    not support it. You will have to manually tailor your particular case.
    If instead you want to split it in a number of sections to adapt the length to your needs
    you can by either:
    process_files(
        language="ell",
        files_dir="my_input_files",
        n_sections=10, # At most 10 sections
    )

    Or if you have a number of seconds per section in mind:
    process_files(
        language="ell",
        files_dir="my_input_files",
        cut_at_seconds=30, # At least 30 seconds
    )
    This modifies texts and audios.

My chapter-structured I mean

    example
    ├── texts
    │   ├── chapter1.txt
    │   └── chapter2.txt
    └── audios
        ├── audio1.mp3
        ├── audio2.mp3
        └── audio3.mp3

Caveats:

  • I was only able to test it briefly. Worked very well except for one cut of a 1h20~ audio where it was 10~seconds off (and then it fixed itself later on). Maybe precision issues.
  • aeneas is (sort of) hard to install.
  • No guarantees on aeneas supporting your language.
  • Some parts are still WIP

The code is in the already linked repo.

3 Likes