Validate segments

[sw1313]

Related issues / PRs. Summarize issues.

Validate segments not empty in subtitle_manager.py

Summarize Changes

Perform defensive checking to avoid the "list indices must be integers or slices, not str" exception caused by null values.
test file: https://github.com/anars/blank-audio/blob/master/10-minutes-of-silence.mp3

[jhj0517]

Hi.
I tried to reproduce it with the file, but it was not reproducible with lastest version.
Can you check?

[sw1313]

_whisper-webui_logs.txt
This is the log of me pulling the latest version files using git and running them.

[jhj0517]

I assume this happens before merging #480?

Can you fetch upstream's and try again?
It leaves a message instead of throwing an error.

[sw1313]

I tested the latest code, but the error still occurs.
I consulted the reason for the error, and the response is:
If the upper layer Whisper reasoning or VAD result is empty, pass the empty list directly to subtitle_manager.
The code logic originally expected result to be a dict with the key "segments" inside.
The actual incoming result is a pure list, or other structure that does not contain "segments".
In the next step in iterate_result or write_result function, when doing result["segments"], if result is a list, then access it using string index ("segments") on the list.
Python will raise a "list indices must be integers or slices, not str" error.

[sw1313]

The issues #468 case1 is an error caused by whisper when no valid data is output for the audio(include audio track).
A more strict illusion penalty must be used to reproduce this error, so that no illusion data is written into the subtitles.
Case3 is an error caused by the failure to obtain the audio track during video detection phase. (case3 has been fixed)

[jhj0517]

I could reproduce it now.
But in my opinion it should not just silently return nothing, but leave a message and then pass the case.

May I edit some?

[sw1313]

Of course, I consulted OpenAI for a solution only after you were unable to reproduce it.

[jhj0517]

Confirmed that it is no longer reproducible