audio_capture.py:
- Move _fft_bands from inner function to method (avoid per-frame closure)
- Pre-allocate channel split buffers and RMS scratch arrays
- Use in-place numpy ops (np.copyto, np.multiply) instead of copies
- In-place FFT smoothing instead of temp array allocation
- Cache loop-invariant values as locals
- Fix energy index to wrap-around instead of unbounded increment
audio_stream.py:
- Pre-compute interpolation arrays (band_x, led_x, full_amp, indices_buf,
vu_gradient) once on LED count change instead of every frame
- Pre-compute VU meter base/peak float arrays in _update_from_source
- Reuse full_amp and indices_buf buffers across frames
- In-place spectrum smoothing to avoid temp allocations
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
alexei.dolgolyov
04ee2e5830