A microphone connected via JST-SH feeds I2S audio into the RP2040, which reads it through a PIO state machine (the RP2040 has no dedicated I2S hardware, but PIO handles the protocol cleanly). The RP2040 enumerates over USB-C to the host as a composite USB Audio Class device with both capture and playback endpoints, so the Taishan Pi sees a standard USB sound card accessible through ALSA. Playback flows in reverse: the host writes audio to USB, the RP2040 streams it as I2S into a MAX98357A Class-D amplifier driving the speaker output. Two microphone ports share the same I2S bus with their L/R pins set opposite, enabling stereo capture or future beamforming without a board respin.

The key decision was routing audio through the RP2040 as a USB device rather than wiring I2S directly to the RK3566's I2S controller. Direct I2S would have required a custom Rockchip device-tree overlay and codec binding, locking the board to specific kernel versions and distros; the USB Audio Class path is fully distro-agnostic and portable to any Linux SBC. The board is kept at 2 layers because voice-band I2S at ~3 MHz BCLK tolerates simpler routing as long as the bottom-layer ground pour is continuous, which saves cost on the prototype run with a clear path to a 4-layer revision if noise testing demands it.