Various audio filters

[Purfview]

Various audio filters:

--ff_dump
--ff_track
--ff_fc
--ff_mdx_kim2 - It's available only in Faster-Whisper-XXL.
--ff_mp3
--ff_sync
--ff_rnndn_sh
--ff_rnndn_xiph
--ff_fftdn
--ff_tempo
--ff_gate
--ff_speechnorm
--ff_loudnorm
--ff_silence_suppress
--ff_lowhighpass

Notes:

Needs FFmpeg.
These filters shouldn't be used in Subtitle Edit lower than v4.0.5, because lower versions of SE are using low quality audio input.
--ff_silence_suppress's internal command can't be longer than 32768 chars [explanation is there].

[ghost]

--ff_rnndn_sh
is thje goat ig

[Purfview]

do you recommend to change it with this
--ff_rnndn_sh --ff_rnndn_xiph --ff_silence_suppress -40 2

I recommend to test it yourself. 😉

On my limited tests I've seen improvement.
Btw, I've seen worse results on audio parts with songs-music when using those rnndn models, but that is expected as singing is not exactly "speech".

[ghost]

alrr will test it thanks for improvements bro

[Purfview]

I've noticed that "sh" suppress singing when "xiph" kept it, generally "sh" is stronger at suppressing various noises.
Btw, "sh" is abbreviation for "somnolent hogwash". [it doesn't mean anything]

[ghost]

@Purfview some podcasts have songs at the beginning, for example transcription didn't have them, I'm quite happy with sh because that's what I want, but first I need a linux build to try xiph :D

[Purfview]

These rnndn filters are very fast but not so good for the today's standards, nothing to write home about when compared to the state of art MDX-Net vocal extraction models.