Debugging Embedded Audio Static / Streaming-PCM Noise

Static on an I2S speaker (MAX98357A, PCM510x, etc.) has several distinct causes that sound similar to the human reporting them, and the wrong ones are seductive because they're about hardware — the amp, the clock, the ground. This skill exists because in the robocar-unified voice bring-up (2026-07) two hardware theories were wrong (uninitialised DMA content; BCLK stop/start pops) and cost several reflash cycles each. The fix only came from measuring the actual samples at each stage. Measure first; theorise last.

The one law

You cannot hear the audio; the user can, and the samples can. Do not reason about the amp/clock/DMA from the symptom word "static." Get the PCM bytes and compute statistics on them — that localises the corruption to a pipeline stage in one pass, with no hardware and no guessing.

Step 1 — get a precise symptom from the user (it discriminates the cause)