The room was wrong, and the session could not be repeated

The situation

A crew from the London Contemporary School of Piano filmed a series inside Blüthner's piano factory: a building made for building instruments, not for recording them. Large, hard surfaced, and generous about giving sound back to you slightly late. The schedule could not slip, and nothing could be filmed a second time.

The decision

One option was to apply the same noise-reduction setting across the take.

We took a longer one. The room noise was isolated onto its own track, aligned in phase and start point with the original recording, and then dosed by automation, decided passage by passage rather than set once.

Why

Because the room is part of this recording. On a channel that teaches piano, it carries the size of the space and the sense of someone playing an instrument inside it. We wanted to reduce distraction without losing that character.

Because the performance moves. During improvisation the posture and speaking level change. We adjusted the treatment with the passage, listening for where it began to affect the voice or instrument.

Because of what the audience comes to hear. For people learning or listening closely to piano, the sound of the instrument is part of the lesson. That made a natural result our priority.

The mechanism

The separate noise track let us adjust the treatment passage by passage. The useful difference was the control it gave us on this recording.

What the result sounds like

A room with a person in it. Which is what it was.

Sources and evidence

What here is established acoustics and what is the choice we made on this project.

What is established. How much of a room you hear is a balance of direct to reflected sound rather than a single level, the relationship the Institute of Professional Sound calls critical distance; and automatic noise reduction lowers that room by trading it against the voice, a cost measured in peer-reviewed speech-enhancement research (Loizou and colleagues) and in more recent work on the artefacts such processing leaves behind. Neither removes a room for free.

Our method. Isolating the room onto its own track and dosing it by automation, passage by passage, rather than setting one threshold across the take, is the choice made here.

References.

• Institute of Professional Sound. Critical distance. ips.org.uk. Room sound depends on direct-to-reverberant balance, not distance alone. Technical reference, not peer-reviewed research.
• Kim, G. and Loizou, P. C. Gain-induced speech distortions and the absence of intelligibility benefit with existing noise-reduction algorithms. pmc.ncbi.nlm.nih.gov. Noise suppression introduces amplification/attenuation distortion; conventional NR does not by itself raise intelligibility. Constraining amplification distortion CAN yield gains; does not establish a viewer-retention effect.
• Multiple authors. Speech enhancement artefacts and naturalness. arxiv.org. Modern enhancement can introduce artefacts; naturalness remains a constraint. Preprint; use as supporting rather than sole evidence.