Circuit to combine two PWM generated sound signals

U7A is surplus to requirements. Note that R9/10 could just as easily go to U1B's +ve input as U7A's.

A capacitor in series with the loudspeaker would be useful, to eliminate the DC bias on it, and so reduce the current required from U1B.

In principle, the diagram has all the bits and pieces in block diagram kind of level and it might even work in practice to output something that resembles audio.

The actual implementation is just very poor and is missing a couple of obvious things it should have, so it might be low in volume and have poor audio quality, at least if compared to what it could be with a proper design.

First of all, at high level, the op-amp isn't strong enough to drive the speaker, and the DC bias is passed on to speaker, which needs to be AC coupled. Technically there should be an amplifier capable of driving the speaker.

Then the actual op-amp and component values to bias the op-amp and the gain it has may not be very suitable for the old LM324. At 5V supply, it will have input voltage range between 0V and 3V only, so bias should be 1.5V. On the other hand, at any appreciable level of current, the output range will be between 1V and 3.25V, so DC bias should be about 2.125V.

Summing two 5V signals together at 1x gain means the output will clip.

It is not meaningful to calculate why the values are not very good, as they need to be recalculated anyway, so it may make sense to just recalculate the values appropriately from scratch, and then check if what AI told you makes sense or not.

As said it will make also sense to AC couple the input, but frankly, the LM324 is so weak that even the MCU PWM pins might drive the speaker with more current, and mixing the audio could be done by just wiring the PWM outputs together with resistors. It may make sense to exchange the LM324 to something newer as well.

There are some weird things such as one opamp used as buffer for reference voltage, but as the reference voltage goes only to one high impedance node, the input of second op-amp, the first op-amp can be removed without any functional difference.

You need to add a blocking capacitor in series with U18's (-) input, otherwise the 1.8V bias volage will be amplified by a factor of 1 to 2, depending upon how many inputs are connected.
A value of 1µF will give a low frequency rolloff below 20Hz.

Have you considered the effect of the low pass filters (R4 and C6, R7 and C7) on your audio?

The given values (10k and 4.7nF) give a cutoff of about 3.4kHz.

• Is that high enough to pass the audio you are generating?
• Is that low enough to filter out the PWM signal?

It should work if you have input coupling capacitors to remove any residual DC level that might swamp the op-amp output. However you won't have much output headroom with an LM324 on 5 volts. It looks like the output should be centred (around 1.8 volts) between the clipping points (0 volts and 3.5 volts) so, if you can live with that then no problem.

Connecting a speaker to the output is a big mistake.