Good evening

I have been trying to solve this puzzle since yesterday… although I use the debugger to analyze the code of python-ffmpegio, I unfortunately did not come to a solution.
Maybe someone with experience with python-ffmpegio knows what the solution must look like....

To explain the task, I have a working filter_complex command for the ffmpeg command line, but I am not able to transpose it to python-ffmpegio, so that Python does not save a file but return the filtered data back as a numpy array:

ffmpeg -i "Stereo.Wav" -filter_complex "[0:a]pan=1c|c0<0.5*c0+0.5*c1[a_mono];[a_mono]dynaudnorm=targetrms=0.95:correctdc=1[a_normalized]" -map "[a_normalized]" "Normalized-Mono.wav"

ffmpeg is doing this:

1. Read a stereo audio file and convert it to mono by using the pan filter
2. afterwards, the ffmpeg dynaudnorm filter is applied to the mono audio data
3. in the end, the normalized mono audio file is saved to disk

The challenge for python-ffmpegio lies in this:

If possible, the filtered result should not be saved to a file, but returned as a numpy array. This way the data can be processed further without having to work with big temporary files.

I can contribute this additional information:

In Discussion Nr. 13, @tikuma-lsuhsc explained, how we can pass an numpy-array to an ff.audio.filter() and get the result back as a numpy-array, so that we do not have to write a file 😀.

He posted this ingenious, working code (please note the 'anull' parameter in the last line which allows to not write the result to a stream):

import ffmpegio as ff
import numpy as np
from matplotlib import pyplot as plt

fs = 16000
T = 1
x = np.random.default_rng().integers(2^-15,2^15,dtype='int16',size=(T*fs,2))
# no explicit filtering, just implicit stereo->mono conversion
fsout,y = ff.audio.filter('anull', fs, x, ac=1) 

Unfortunately, I did not manage to:

• add a filter_complex parameter to the example code.
  For example, I got this error message, if I just added a filtergraph: Multiple -filter, -af or -vf options specified for stream 0, only the last option '-filter:a anull' will be used.
• Additionally, the option expr='anull' must not be empty for ff.audio.filter(), so we can't just pass a filter_complex parameter.
• Finally, I do not know how we could combine the expr='anull' parameter into this definition:
  -filter_complex "[0:a]pan=1c|c0<0.5*c0+0.5*c1[a_mono];[a_mono]dynaudnorm=targetrms=0.95:correctdc=1[a_normalized]"

Many thanks for any ideas / hints.
Also to learn if it is a limit of python-ffmpegio and can not work, then that is already a big help :-)

Thanks a lot, kind regards,
Thomas