Result

Motivation

Modern speakers require a lot of DSP magic to sound as good as they do.
Speakers traditionally needed to be built very carefully to achieve a very flat frequency response and as few artifacts as possible.

These days, many mobile devices like phones, laptops, etc. do a lot of digital signal processing in software in order to make tiny speakers output a lot of sound.
Many different tricks are used, such as psycho-acoustic bass enhancement, limiters (allowing for higher power peaks) and volume-dependent equalization (frequency response of human hearing is very volume-dependent).

The Asahi Linux project (Linux on Apple Silicon MacBooks) has done a lot of work to make MacBooks sound as good as possible on Linux:
AsahiLinux/asahi-audio: Userspace audio for Asahi Linux

How?

Using Room EQ Wizard the frequency/impulse response of the built-in speakers was measured.
Even with very suboptimal measurement equipment (cheap microphone, questionable audio interface), we can clearly see the sloped bass response (expected with speakers of this size) and then also a very noticeable peak/resonance at ~4kHz.
This peak is very audible and leads to a harsh, distorted sound when listening to music.

REW can do arithmetic operations on the curves, so after generating a filter curve for a 300Hz slope, both channels were divided (A/B) against that curve and then inverted (1/A) and exported as a .wav impulse response.
That impulse response .wav file can then be used in a convolution DSP filter.

The DSP configuration/pipeline of a 14" MacBook Pro was then used as a template, the multi-speaker setup (modern MacBooks have 6 speakers!) was stripped-down to just a single stereo pair and the parameters were changed a bit.
The impulse response .wav was (of course) replaced with our newly generated REW output.

Links