The Lucier Machine — User Guide

Lucier-inspired iterative room filtering: one synthetic room impulse response is applied repeatedly so that its resonant spectral fingerprint progressively dominates the source.

Author: Shai Cohen Affiliation: Department of Music, Bar-Ilan University, Israel Version: 0.5.1 (2026) License: MIT License Repo: https://github.com/ShaiCohen-ops/Praat-plugin_AudioTools

Contents:

What this does Quick start Presets Parameters Synthetic room model Iterative filtering Output & visualization Historical, technological & compositional context

What this does

The Lucier Machine repeatedly filters a selected Sound through one fixed synthetic room impulse response. The room model contains a direct arrival plus a stochastic field of decaying reflections. Because every iteration uses the same impulse response, frequencies favored by that room transfer function are reinforced again and again while other spectral regions become comparatively weaker.

The result is a progressive transformation from source identity toward a ringing spectral pattern determined jointly by the source and the synthetic room. The processor is designed around the principle associated with Alvin Lucier's I Am Sitting in a Room, but it does not simulate a physical room geometrically and it does not reproduce the original tape-recorder performance procedure.

Core idea: within one run, the room fingerprint stays fixed. Randomness is used only when that impulse response is created. The same room is then applied on every pass, which is what allows its resonances to accumulate coherently.

Quick start

Select exactly one mono or stereo Sound.
Run The_Lucier_Machine.praat.
Start with Lucier-style (30 iterations) to hear the central process.
Use Quick Preview for a faster transformation check, or Extended Transformation for a stronger spectral takeover.
For Custom, set the room duration, RT60, reflection density, direct/room energy balance and number of iterations.
Leave Draw_visualization enabled if you want to compare the source spectrum, final spectrum and measured spectral emphasis.
The result is named original_lucier_Preset.

Presets

Preset	IR duration	RT60	Reflections	Pre-delay	Direct energy	Passes
Lucier-style	1.5 s	1.0 s	1000	10 ms	92%	30
Quick Preview	1.0 s	0.8 s	500	10 ms	88%	10
Extended Transformation	2.0 s	1.2 s	1500	12 ms	94%	50
Small Room	0.8 s	0.4 s	600	5 ms	90%	30
Large Hall	3.0 s	2.5 s	2000	25 ms	85%	30

All built-in presets use a per-pass peak target of 0.95. Draw_visualization and Play_result remain user controls. “Lucier-style” means inspired by the iterative resonance principle; it is not a historically exact reconstruction of Lucier's equipment, room, or recording.

Parameters

Parameter	Default	Behavior
IR_duration_s	1.5 s	Total duration of the synthetic mono room impulse response. It must be longer than two samples at the source sampling rate.
RT60_s	1.0 s	Controls the exponential amplitude envelope of the stochastic reflection field. The envelope reaches 0.001 of its initial amplitude, or -60 dB, after one RT60.
Number_of_reflections	1000	Number of randomly timed reflection events. Because reflection energy is normalized independently of this count, increasing the number primarily changes density, not the direct/room balance.
Pre_delay_s	0.01 s	Location of the direct arrival. Random reflections begin at least one sample later. The pre-delay is removed again after each convolution pass, so it does not accumulate from iteration to iteration.
Mic_proximity_gain	0.92	Despite the name, this is interpreted as the direct-path energy share, clamped to 0–1. A value of 0.92 requests approximately 92% direct energy and 8% reflection energy before the final common IR-energy normalization.
Number_of_iterations	30	Number of virtual playback/re-record filtering passes. The script limits this to 100.
Per_pass_peak_target	0.95	True peak normalization after every pass, clamped to a maximum of 0.99. This is level stabilization, not an attenuate-only safety ceiling.
Draw_visualization	yes	Builds the multi-panel diagnostic view described below.
Play_result	yes	Plays the final result after processing.

Synthetic room model

The script first creates a mono impulse response. This single room response is used for every source channel and for every iteration.

Direct path and reflection field

The direct arrival is placed at Pre_delay_s with an amplitude derived from the requested direct-path energy share.
Reflection times are drawn randomly between just after the direct arrival and the end of the IR.
Each reflection receives a Gaussian random sign/amplitude and an exponential RT60 decay.
The reflection field is scaled so that its total discrete energy matches the requested room-energy share, independently of reflection count.
After the direct path is added, the complete IR is normalized to unit discrete energy. There is no arbitrary IR peak normalization.

reflection envelope = exp(-t × 6.907755 / RT60) Direct energy share = Mic_proximity_gain Reflection energy share = 1 - Mic_proximity_gain Final room IR: direct path + normalized stochastic reflections Then normalize complete discrete IR energy to 1

No random seed is exposed. Running the same settings again creates a different synthetic room fingerprint. Within a single run, however, the generated IR remains fixed for all passes.

Iterative filtering

Repeated convolution through one fixed room

Praat convolves the current Sound with the mono room IR. For stereo input, the same IR filters both channels; the channels are not folded to mono. After convolution, the script extracts a segment beginning at the direct-arrival time and exactly as long as the original source. This re-aligns the pass to time zero and prevents pre-delay from accumulating.

Ideal frequency-domain picture: X_N(f) = g_N × X_0(f) × H(f)^N X_0(f) = source spectrum H(f) = fixed synthetic room transfer function N = number of passes g_N = scalar level stabilization

The scalar peak normalization cannot change spectral shape; the repeated factor H(f) is what produces progressive resonance reinforcement. However, the implemented process is not exactly H(f)^N: after every pass the convolution tail is truncated back to the original source duration. The visualization therefore compares the ideal room emphasis with the emphasis actually measured in the output instead of assuming they are identical.

Per-pass level stabilization

After every extraction, the current Sound is peak-normalized to Per_pass_peak_target. Stereo uses one common gain factor, preserving the left/right level relationship. This stabilization keeps successive generations numerically manageable but is an explicit part of this digital model rather than a neutral recording-history simulation.

Direct energy = 100%: the room IR contains only the direct impulse, so there is no room-resonance transformation. The waveform shape is essentially passed through the aligned impulse, but the per-pass peak normalization can still change its overall level.

Output and visualization

Input: exactly one mono or stereo Sound. Other channel counts are rejected.
Channel count: preserved.
Sample rate: preserved.
Duration: restored to the original source duration after every pass and in the final output.
Time origin: the working signal is rebased to 0; the final output therefore starts at 0 rather than retaining a non-zero source start time.
Output name: original_lucier_Preset.
Wet/dry: there is no separate wet/dry control. The direct/room energy balance is built into the room IR itself.
Final peak: for a non-silent processed result, the last pass has been peak-normalized to the selected per-pass target.

What the visualization shows

Original waveform and final-pass waveform, each drawn with an explicit symmetric amplitude range.
Spectrum before and after on one shared logarithmic frequency axis. Mean level is removed so the panel emphasizes spectral shape rather than gain.
Room fingerprint H(f), displayed as a log-spaced band-averaged curve rather than an unreadably dense raw stochastic spectrum.
Emphasis per pass: the ideal room-shape curve compared with the measured (X_N - X_0) / N spectral emphasis.
Measured / ideal emphasis ratio, reported as a diagnostic rather than treated as proof of a single cause.
Reinforced-frequency markers derived from the measured source-to-output emphasis. These markers therefore reflect frequencies the source actually excited, not merely peaks of the room response in isolation.
Process diagram showing source → fixed room → level stabilization → repeated room filtering → output.

For stereo sources, the waveform and spectral display uses whichever source channel has the larger peak, and shows the corresponding result channel. It does not create a mono sum for display, avoiding phase cancellation in the visualization.

Historical, technological and compositional context

Alvin Lucier and I Am Sitting in a Room

Alvin Lucier's I Am Sitting in a Room dates from 1969 and was conceived for voice and electromagnetic tape. The original score specifies a simple electroacoustic chain: one microphone, two tape recorders, an amplifier and one loudspeaker. A spoken statement is recorded, played back into a room, recorded again, and the new generation is repeatedly recycled through the same space. With each generation, the room's resonant frequencies are reinforced until the intelligibility of the speech gives way to sustained, ringing spectral patterns.

The spoken text is not merely narration placed on top of the piece: it describes the process that generates the music. Lucier also frames the procedure as a way of smoothing irregularities in his speech. This self-referential design joins acoustic experiment, autobiography and compositional form: the material explains the system, then the system gradually transforms the material that explained it.

Technology: from tape recursion to digital recursion

The 1969 procedure depended on physical playback and re-recording. Loudspeaker response, microphone response, tape electronics, noise, placement and the architecture of the room all belonged to the feedback path. Later realizations preserved the same recursive principle with digital recording technology; for a 2014 MoMA performance, Lucier and James Fei used a laptop in place of the two tape recorders.

From a signal-processing perspective, an approximately linear, time-invariant room can be represented by an impulse response h(t). Convolution with that impulse response models one passage through the room, and repeating the same operation produces the idealized spectral relationship H(f)^N. The Lucier Machine makes that abstraction explicit: it replaces the physical microphone–room–loudspeaker loop with a generated impulse response and repeated offline convolution.

Compositional significance

I Am Sitting in a Room is a landmark of process-based experimental music because the audible form is not produced by writing a sequence of pitches in advance. The composer defines a procedure and allows the acoustic properties of a space to determine the detailed result. The room therefore ceases to be a neutral container for music and becomes an active spectral agent.

The process also reverses the normal relationship between signal and environment. At first, the voice dominates and the room is barely noticed. Through repetition, the room becomes increasingly audible until its resonances become the principal musical material. What begins as speech moves toward a harmonic field while retaining traces of the original temporal articulation.

Relation to this AudioTool: this script models that compositional principle, not the historical performance itself. It uses a synthetic stochastic room IR, one fixed transfer function per run, direct/room energy normalization, direct-arrival re-alignment, source-length truncation and per-pass peak normalization. A physical Lucier realization instead sends each generation through an actual acoustic/electroacoustic chain. The two share the logic of repeated resonance reinforcement, but their material conditions and therefore their results are different.

Sources and further reading

Museum of Modern Art. “Collecting Alvin Lucier's I Am Sitting in a Room.” Historical account of the 1969 work, the resonance process, and Lucier's later transition from analog tape to a laptop-based realization.
Museum of Modern Art. Alvin Lucier, I am sitting in a room, 1969/2014. MoMA collection record.
DRAM. Notes for Alvin Lucier: I am sitting in a room. Describes the two-tape-recorder process and the progressive transformation of speech into room resonance.
Wesleyan University. Lucier's 90th Birthday Celebrated with 90 Artists Recording I Am Sitting in a Room. Context on the work's continuing performance history and the way each realization is shaped by its particular space.