Audio Source

Audio Source captures live microphone / input audio and exposes it as continuous Signals. It is a pure source node: it has no inputs and no user-facing parameters. On each evaluation it reads per-channel capture handles from the engine's MEMORY_REGISTRY and emits the most recent trailing samples as the left, right, and signal outputs.

The node is non-blocking and never errors. If a capture handle is missing, idle, or has not yet produced samples, the corresponding output is simply an empty Signal (length 0). The signal output is an alias for the first non-empty channel, giving a convenient mono tap that works whether the device is mono or stereo. The optional window_size runtime parameter, when bound by capture control, limits each output to the last $N$ samples of the ring buffer.

Sample values are dimensionless normalized audio amplitudes (typically $x\in [-1,1]$); they carry no physical unit and no uncertainty ($\sigma =0$), since the samples are raw device readings rather than derived quantities. Attach scale_offset downstream to map to volts or pascals with an explicit unit.

Although it draws from an external, time-varying capture buffer, the node itself declared as stateless: it holds no internal accumulator between evaluations. All persistence lives in the shared capture ring buffer referenced by the runtime handle IDs (left_handle_id, right_handle_id, handle_id).