St. Anna, Bia
Tools and equipment
A wireless microphone pair I already had, a Windows computer with Audacity, LoopAuditioneer for the loops and Hauptwerk 4 for playing, and processing programs written with AI help.
This text was generated by AI from curated notes and may contain errors. Corrections welcome
For the recording
| Tool | What I used it for | My verdict |
|---|---|---|
| BOYA BOYALINK (opens in a new tab) | The dual wireless system I already had: its two transmitters as a stereo pair, about 50–60 cm apart, and the receiver in stereo mode, connected by USB. | Not suitable for sampling sustained pipes; see difficulties. |
| Windows 11 computer | The recording computer. | Its voice processing and audio enhancement had to be switched off. |
| Audacity (opens in a new tab) | Recording at 48 kHz in stereo, as 32-bit float. | Also used for editing; see below. |
I switched off all the selectable processing: the BOYA (opens in a new tab) noise cancellation (steady blue lights on both transmitters), Windows voice processing and Windows audio enhancement. Speech processing would damage the attacks, the bass and the reverberation. That still did not make the microphones neutral.
For editing and playing
| Tool | What I used it for |
|---|---|
| Audacity (opens in a new tab) | One region label per pipe, then blower-noise reduction with the room tone as the noise profile; export as 24-bit WAV. |
| LoopAuditioneer (opens in a new tab) | Two long sustain loops per pipe, chosen by ear, a release cue, and the measured pitch, written into the WAV file. |
| Hauptwerk 4 | Playing the sample set, on a virtual console that follows the real one. |
| Python and NumPy | The processing programs for the finishing work of 25–28 October 2026, written and run by ChatGPT/Codex. |
What the AI did, and what it did not
ChatGPT/Codex wrote and ran the processing programs, in Python and NumPy, and edited the Hauptwerk files. The audio work was ordinary signal processing: no generative audio and no neural noise reduction. I supplied the recordings, heard the problems, made the decisions, tested everything in Hauptwerk and went back to backups when something did not work.
The programs change only the audio bytes they are meant to change and copy everything else in the WAV file unchanged, so loops, release cues and pitch data survive. The left and right channels always get the same processing.
Generative image editing was used only for decoration on the virtual console: a gold crest and two photo insets made from my photos. They are not exact copies; for inscriptions and details, see the original photos.
For finding the microphone problem
To find out where the loud starts came from, I played the pipe organ voice of a Yamaha YDP-165 digital piano at home and recorded it through the BOYALINK (opens in a new tab), both on a computer and on a phone. Both times the tone dropped by about 5–8 dB after about one second, while the phone's own microphone stayed steady.
The recording timer
The project's recording timer is a full-screen cue display for the organist that alternates timed “silence” and “hold note” phases. In my first session I used its default of 10 s for both phases: 10 s of sound, 10 s of silence. That turned out to be more than needed; 10 s of sound and 5 s of silence is enough, so for the next session set the silence phase to 5 s.
Open the recording timer (opens in a new tab) (in English).