The Portable Freeware Collection Forums

I have an audio file where an annoying segment of few seconds is repeated at different times in the file.

I want to feed a copy of this annoying segment to a tool to recognize all instances and remove them, but I couldn't find a way.

This seems like a simple enough idea. Is there really no way to do that in Windows?

vevy wrote: ↑Sun Feb 11, 2024 11:02 pm I have an audio file where an annoying segment of few seconds is repeated at different times in the file.

Great question, I don't even know how I would find a similar audio element inside a file, much less delete multiple instances.

Suggestion (manual method, not very sophisticated):
Use your favorite audio editor e.g. Audacity ( https://www.portablefreeware.com/index.php?id=648 ), locate, select and cut the unwanted parts of the audio file.
Then export the resulting audio file.

webfork wrote: ↑Mon Feb 12, 2024 6:55 pm Great question, I don't even know how I would find a similar audio element inside a file, much less delete multiple instances.

That's what I though must have existed already. It seems easily doable without AI or such.

@juverax

Thanks for the tip, but I really don't want to do it the tedious way. That's a computer's job if you get what I mean.

https://audioz.download/software/win/ has many audio processing tools.

Vevy said: "I have an audio file where an annoying segment of few seconds is repeated at different times in the file"

Question: Is this "annoying segment" exactly the SAME, I mean the sequence of bytes encoding this segment is repeated at different places in the audio file?

1 - If YES, then IMO it is relatively trivial to write an app that parses the RAW audio file, identifies the identical sequences and deletes/replaces them with something else.
Note that conversion between a raw file and a compressed file may introduce spurious bytes.
However the problem is formally similar to "finding a string inside a string" which is a classic algorithm, and there are many web sites where you could look for a solution:
https://www.codeproject.com/
https://www.geeksforgeeks.org/
https://stackoverflow.com/ . . . etc.


2 - If NO, well it depends whether these annoying segments occur at regular intervals - even if the spectrum is slightly different - then all it takes is to remove the bytes at the appropriate positions. Consult the web sites mentioned above, and alike.


3 - If the "annoying segments" are not identical and occur at different places in the audio file, then IMO this is a more difficult problem, because technically you don't have a simple way of "specifying" the annoying segments. Maybe this is a problem for AI enthusiasts .
Once, I had a similar problem: I was trying to transfer an old 78 rpm vinyl disk that was badly scratched. Each peak caused by the scratch had a slightly different envelope and occurred at slightly different intervals.

Thanks a lot for the detailed reply juverax

The answer is yes. It is the same recording inserted every 7-8 minutes (interval varies a bit). The same recording, but I don't know if it will be byte-perfect. There are a lot of image deduplication tools that work even on similar, non-identical images. I just can't find one that does the same for audio signal.

Is the segment the same length? If so you would "only" have to identify the starting timestamps e. 6.58 13.23 etc and then use a (cli) tool to split the file in to sections, say the fragment is 10 seconds:
0-6.58; 7.08-13.23; 13.33-... and join them back
still requires manual processing of course

I don't know what file it is but perhaps if you can find a audio-to-text program, and if the repeating section has a clear identifiable word in it e.g. "You're listening to TPFC Radio" that might help you automate finding the timestamps.

lintalist wrote: ↑Sat Feb 17, 2024 12:55 am Is the segment the same length?

The annoying segment is the same length (same recording), but occurs at variable intervals.

I don't know what file it is but perhaps if you can find a audio-to-text program, and if the repeating section has a clear identifiable word in it e.g. "You're listening to TPFC Radio" that might help you automate finding the timestamps.

That's a great idea! I have speech-to-text software that can output to srt.

That gets us to semi-automation. Getting closer!