Linear loudness normalization in FFMPEG is in fact dynamic

Question

I followed this post to normalize the loudness of a clip. I first measure the loudness:

$ ffmpeg -i "Take109 000-012.wav" -filter:a loudnorm=I=-24:TP=-9:LRA=5:print_format=summary -f null -
...
[Parsed_loudnorm_0 @ 0x7fbfc542c440] 
Input Integrated:    -16.5 LUFS
Input True Peak:      +0.0 dBTP
Input LRA:            18.0 LU
Input Threshold:     -27.5 LUFS

Output Integrated:   -20.2 LUFS
Output True Peak:     -9.0 dBTP
Output LRA:           12.9 LU
Output Threshold:    -31.0 LUFS

Normalization Type:   Dynamic
Target Offset:        -3.8 LU

I then feed those values into the linear normalization command:

$ ffmpeg -i "Take109 000-012.wav" -filter:a loudnorm=I=-24:TP=-9:LRA=5:measured_I=-16.5:measured_TP=+0.0:measured_LRA=18.0:measured_thresh=-27.5:offset=-3.8:linear=true -ar 48000 "Take109 000-012-norm.wav"

The problem is that the result has dynamic normalization with the second part of the sentence significantly quieter than the first, probably due to a spike at the very end of the sentence.

Here are the samples: input and output, with voice starting at 00:05.

How can I use linear loudness normalization, in FFMPEG or other, with a target loudness of -24 LUFS for example?

Answer 1

As mentioned in a comment, FFMPEG will revert to dynamic in the conditions specified in the manual:

linear

Normalize by linearly scaling the source audio. measured_I, measured_LRA, measured_TP, and measured_thresh must all be specified. Target LRA shouldn’t be lower than source LRA and the change in integrated loudness shouldn’t result in a true peak which exceeds the target TP. If any of these conditions aren’t met, normalization mode will revert to dynamic.

In my case, a linear normalization with those settings would result in a soundwave going above 0 dB, so the normalization became dynamic.

I found this in an experiment of not using a limiter in sound pre-processing. I learned from the experiment and re-enabled the limiter.