Numbers to a voltage

Sound is a voltage that moves

A headphone's driver moves with the voltage across it, and the air moves with the driver. Digital audio is that voltage written down as numbers, one every few tens of microseconds: a sample. Playing it back means turning each number into a voltage again, at the same rate and in the same order.

That is the job of a DAC, a digital-to-analog converter.

The ESP32-S3 has none

The ESP32-S3's pins are digital: each is HIGH or LOW, and none can hold a voltage in between for a sample. What it does have is I2S, a peripheral that sends samples out as bits on three wires, with the timing kept by hardware rather than by your sketch. The TK103 is on the other end of those wires.

The chip maps its largest number to 2.1 V RMS, the full-scale output TI quotes. For a sine that just reaches full scale, that is about 2.97 V at each peak (worked out). Negative numbers make negative voltages: a charge pump inside the chip makes its own negative supply, about −3.3 V, so the output swings above and below ground. Zero is 0 V, and no capacitor is needed between the chip and the jack to block a middle voltage.

What the chip brings

• Two channels. Left and right, from one stream of numbers.
• Low noise. TI gives a typical dynamic range of 112 dB for the PCM5102A: the gap between the loudest it plays and its own hiss, measured at 48 kHz with 24-bit samples.
• Most formats. 16, 24 or 32-bit samples, from 8 to 384 kHz. On this board, with no master clock wired, only some of those combinations work, and Three wires, no master clock says which.
• Line level. The output is the signal a powered speaker expects at its input, not the power a bare speaker needs: Line level.