dsp-research-lab

A Claude Code workspace for audio DSP research — algorithm exploration, objective measurement, and accumulation of a validated, reusable building-block library. Modeled as a professional audio R&D department, not a plugin shop: there's no version number, no release cycle, no shipping deadline. The output is validated knowledge, not a product.

This is the research counterpart to a separate plugin-buiding workflow repo Findings here can graduate into that repo's DSP_Modules-equivalent when a real plugin project needs them — see CLAUDE.md § 11.

Philosophy

Research first. Products second. Every DSP algorithm is a hypothesis until measured.

• Objective measurement and subjective listening impressions are always kept separate

and clearly labeled — never blended into a single unlabeled claim

• Nothing is deleted, even failed experiments — a documented dead end saves someone

(including future you) from re-running it

• Promotion to the permanent module library is a deliberate gate, not automatic

TestHarness Scripts

Five working, self-tested measurement tools in TestHarness/scripts/ — each ships with a --self-test flag that validates the script against known-answer reference signals before you trust it on real data:

Plus null_test.py for differential A/B comparison and render_harness.cpp for offline C++ timing without a plugin wrapper.