Use peqdb/autoeq-c for AutoEQ

[audiopwn]

Replaces the JS-based heuristic AutoEQ algorithm with peqdb/autoeq-c, an AutoEQ implementation written in C compiled to WebAssembly. On my M4 MacBook Air the old algorithm typically takes ~850 ms while the new one completes in ~25 ms which is a ~35x speedup.

The previous implementation used a two-stage heuristic approach, searching for candidate filters based on peak/dip detection, then iteratively tweaking their parameters. While conceptually sound, this method was both slow and not very accurate in practice. It also did not feature any sophisticated method to avoid equalizing rig artifacts; if allowed to go above 6 kHz, the 8 kHz peak is often equalized down, and the dip around 10 kHz will be equalized right up to the target which is undesirable.

The new implementation directly minimizes the error between target and current frequency response using a gradient descent based optimizer. We compute exact analytical gradients wrt. every parameter (center frequency, gain, Q factor). This makes the algorithm fast and precise. The new algorithm includes rig-aware adaptive smoothing, with hand-tuned smoothing presets for in-ear and over-ear measurements.

autoeq.mp4