Offline test improvements #150
Conversation
this should mostly not make a difference, but feels semantically more correct
There are two issues here: 1. `line_size` may contain padding (didn't happen in my tests) 2. from: https://git.ffmpeg.org/gitweb/ffmpeg.git/blob/2b5f000d3f6f9e737e918a5438e6c881f65e70e2:/libavutil/frame.h#l405 > For audio, only linesize[0] may be set. For planar audio, each > channel plane must be the same size.
This kind of behaves like libobs, where each chunk of audio is inspected individually by VAD/whisper, until processing of either takes longer than the window length, in which case audio continues to stream in
palana
force-pushed
the
offline-test-improvements
branch
from
59800a1 to
6e5a8af
Compare
palana
force-pushed
the
offline-test-improvements
branch
from
6e5a8af to
e9581f3
Compare
|
|
||
| // sleep up to window size in case whisper is processing, so the buffer builds up similar to OBS | ||
| auto now = std::chrono::system_clock::now(); | ||
| if (false && now > max_wait) |
There was a problem hiding this comment.
kind of undecided if this false should be a parameter, or whether the max wait should be removed
for running tests deterministically between machines you currently want to always only feed a single chunk of audio into gf->input_buffers
honoring max_wait gives you something that is closer to the "plugin within obs experience", i.e. while whisper inference (etc) is running (which takes longer than the wait time on my machine), audio buffers continue to fill up, so on the next run of the whisper loop a bigger chunk of audio is fed into VAD
palana
force-pushed
the
offline-test-improvements
branch
from
6414214 to
bb73bcf
Compare
| std::time_t now_time_t = std::chrono::system_clock::to_time_t(now); | ||
| std::tm now_tm = *std::localtime(&now_time_t); |
There was a problem hiding this comment.
think we don't need this anymore right?
There was a problem hiding this comment.
I kind of like having both timestamps available, local time to orient myself on when a particular run happened and "running time" to compare relative timing within a run
These changes make the offline test deterministic in my testing, i.e. the same input file produces the same
segment_labels insegments.json