The latency of the dataflow is made of the audio framework and the filter you mention in your dataflow.
Regarding the EQ and compressor algorithms, we suggest you to use a FIR-graphic-equalizer and/or mdrc audio-kit algorithms if possible in your use case.
The FIR-graphic-equalizer is an equalizer in which you can choose the number of bands, note that default setting for firType is "linear phase" which adds a firSize/2 samples latency but which has the best behavior in terms of signal quality (for music experts); if your main concern is latency, choose "minimal phase" for firType, it has also good quality performance at the expense of a variable latency (very low, comparable to IIR filtering) depending on signal frequency.
mdrc is a Multi-band Dynamic Range Compressor, it is a 1 to 5 bands Dynamic Range Compressor, when there is more than 1 band, bands separation is made through IIR filtering thus with very low latency and each band may have its own compression curve; maybe it can comply to your requirement alone (without additional equalizer) since you can apply a different compression curve and gain for each band.
Regarding the audio framework,
on STM32N6 we could get very low latency (it works fine up to 0.125 ms), with a computation power extracost. I don’t have the measures on STM32H5 yet, 0.5 or 0.25 ms is probably around the limit on that one.
If you want to experiment with your dataflow, you will find more info in audio-kit reference manual, chapter "How to reduce latency".
