Automatic timing for subtitles by audio analysis

Question

Subtitles

I work with foreign audio books. Like creating subtitles, but the specifics are different.
1) the text is already there - you just need to break it;
2) there are no extraneous noises (as in films);

My task would be ideally solved if I find a program that, similarly to Adobe Audition, will have the “find and mark phrases” function.



Let's say that all sections of silence are more than 2 seconds, so that they are perceived as a border between phrases. As a result, I need a series like this:

0:02:22.39,0:02:23.41
0:02:24.07,0:02:26.66
0:02:27.46,0:02:28.52

while I get it with handles in Aegisub. Can anyone recommend a program to automate this?

Answer 1

You can use the aeneas program under Windows (there are also under Linux, Mac) from the command line.
There is a ready-made aeneas installer at: https://github.com/sillsdev/aeneas-installer/releases
(as of 04/25/2020 includes: eSpeak 1.48.04, FFmpeg 4.2, Python 3.7.4, BeautifulSoup4 4.8.0, lxml 4.4. 1, numpy 1.17.1, aeneas 1.7.3.0)
Command line execution example:
python -m aeneas.tools.execute_task audio.mp3 text.txt "task_language=eng|os_task_file_format=srt|is_text_type=plain" subtitles.srt
where:
audio.mp3 - the name of the input audio file (if without a path, it should be in the folder with aeneas installed)
text.txt - name of the input text file (if without a path, it should be in the folder with aeneas installed). The text must be presented in UTF-8 encoding as a sequence of fragments. The fragments correspond to subtitle lines (that is, you need not continuous text, how many text fragments start with a new line - so many subtitle lines will be )
task_language=eng abbreviated name of the language (es - Spanish, eng - English, etc.)
os_task_file_format=srt file format
subtitles.srt name of the output subtitle file (if without a path, it is uploaded to the folder with aeneas installed)
Note: there is an online version of aeneas at: https://aeneasweb.org/ submit_task
Online requires registration and an e-mail address to which the results are sent.