The reason for the treble rolloff is that engineers are afraid to deviate from antique white papers that prescribe a 1st order 20kHz low pass filter. That a short with a small cap. But that still takes a few dB off from the top octave. This is plain stupid because this is a fixed analog filter for 44k even if you play 96k. And what is worse: you only need it to filter aliasing tones caused by 0dB max level hf measuring sweeps. You will never encounter these peaks in music. Or real life sounds. I cut it out altogether and now it sounds open and airy. Not exaggerated, not attenuated. No pre-ringing. Just a little bit of post ringing which is no problem because it's common irl.
So the output caps are 1st order high pass, which is no problem at 20Hz, but the caps change the sound. Makes it softer, smears in time and veils the sound. If you remove them the music signal (let's call it a test sinus) moves from between - 1.5V <>1.5V (=0V average) to 1.5V<>4.5V (=3V average). That's an offset or bias (like in class A tube amps). That's not playing it 'by the book'. But what the book doesn't say is that the first thing any amplifier does after the volume pot is again the same thing (usually input caps). So the only repercussion is some crackling on the volume pot if you turn it (some static ard because of the 3V). Why do it twice if it hurts the sound? is my question.
All you need for the TDA1543 chips to work is a resistor short between + and - Now this varies with the number of chips because they are in parallel (1/R+1/R+1/R+1/R=1/Rtotal). Iirc the value you need for 8 chips is 220 ohms (it's 2k2 pp). But here is the secret for tuning those piano and female voice peaks, ie prevent oversaturation. If your value is to low it will saturate, if it's to high your volume goes down (like no short > no go). So this is where it often goes wrong. Engineers want lots of volume to get those basses pumping and fast dynamics. But it hurts those sustained midrange peaks like in piano or sopranos. Use your ears not the manual.
So all that you need is to find the point from leg 6 (left output) and 8 (right), probably the same point as before the 1cm metal capacitor, and take that to the output socket. Then find the I/V resistor and measure it and find some 0.25W carbon resistors (simple, small, cheap but good for sound) around that value and experiment. Simply connect output to ground. Just make sure the original is not in parallel or your value is going to be off.
Now this is from memory, and it's been a while, but I reduced the value just 1 tick on my 4x tda1543 from 560 to 470. So half that would do. Don't be nervous, it's not going to hurt the chips, just AB the sound.
I could be mistaken about that value. It could be it is already pretty optimal. But I'm not going to open it up to check it now.
*2 minuutes laterrr*
It's 225. I can measure from the outside. Duh. My memory is good though