Attempt 3 listening: audio-to-audio de-click

Six 10-second excerpts, each from a session the model never trained on. Press play, then switch versions while it plays. The switch is instant and keeps your place, so you can A/B on one spot. Keys 1 to 8 switch versions on the card you last touched, and the space bar plays or pauses it.

A is the raw audio, B is the model at 1×, 2× and 3× strength, and C is your final. D, E and F are difference tracks boosted so they're audible: D is what you changed, E is what the model changed, and F is what the model missed (B1 minus C). Where a difference track would have clipped, the boost was lowered, and the label shows the actual amount. The audio is lossless FLAC.