Repairing Real Audio Capture

The earliest version of RiffCatch was proven with synthetic audio. Generated tones with known pitches and durations went in; notation came out; the pipeline could be tested end to end without a musician in the room. That was the right way to build it, and it was also a comfortable place to stay too long.

This cycle was about leaving that comfort. RiffCatch now records from a real microphone: device selection, input levels, and the unglamorous plumbing between the operating system’s audio stack and the transcription pipeline.

Real input immediately exposed assumptions the synthetic path had hidden. Synthetic tones start exactly when they claim to; a played note has a percussive attack, sympathetic resonance, and room noise around it. Synthetic audio is perfectly level; real recording levels depend on the interface, the instrument, and how hard the player digs in.

The honest status: the microphone path works, and captured performances flow through to notation. What is not yet proven is transcription quality on a real guitar across playing styles, positions, and noise conditions. That calibration work is its own project, and it is the current focus. Until it holds up, we are careful to describe transcription as under test rather than solved.

The lasting principle from this cycle is that audio is evidence. Whatever the transcription does, the recording a Rifflet keeps is the ground truth of what was played. The notation can be corrected later; the moment cannot be re-performed.