Speaker Sim

Real loudspeakers, in a real room, on your headphones.
Website · Downloads · Version française

Speaker Sim turns a pair of measured loudspeakers into a virtual audio output. It combines their CEA-2034 response, a simulated room built from real materials and acoustic treatment, a KEMAR head (HRTF) and an AutoEq correction for your exact headphone model. On Linux it runs live as a PipeWire filter. On Windows and macOS it exports the filters for a system-wide convolver.

Applications ──► [Speaker Sim] (virtual output, becomes the default output)
                   │  4 convolutions: left/right speaker → left/right ear
                   ▼
                 your headphones (physical output)

Install

Platform	Mode	How
Linux	live virtual PipeWire output	curl -fsSL https://devfrp.github.io/SimulSpeaker/install.sh | bash
Windows	export → Equalizer APO	SpeakerSim-Windows-x64.exe
macOS	export → any convolver	Apple Silicon · Intel

The Linux script needs no root. It installs into ~/.local/share/devfrp-speaker-sim and adds the speaker-sim command plus a menu entry. Options (after bash -s --): --uninstall, --purge, --ref v1.0.0, --no-desktop. Re-run the same command to update. The Windows and macOS binaries are unsigned. On Windows: SmartScreen → More info → Run anyway. On macOS: right-click → Open the first time.

From source

System packages (Arch / Omarchy):

sudo pacman -S --needed pipewire wireplumber libmysofa python-numpy python-scipy \
    python-requests python-pyqt6

Debian/Ubuntu: sudo apt install pipewire wireplumber libmysofa1 python3-numpy python3-scipy python3-requests python3-pyqt6.

Or in a virtual environment (libmysofa stays a system package):

cd speaker-sim
python3 -m venv .venv && . .venv/bin/activate
pip install -r requirements.txt
python3 main.py

PySide6 works in place of PyQt6 (automatic fallback, see qt_compat.py).

Features

• Headphones: search 8,850 AutoEq measurements (index pinned at commit 7ae0f56d). Each correction is downloaded once, then cached. Import your own: an AutoEq / Equalizer APO EQ (ParametricEQ, GraphicEQ) used as is, or a raw frequency, dB measurement (REW, Squiglink, AutoEq CSV) corrected toward an AutoEq target (Harman over-ear 2018, Harman in-ear 2019, diffuse field ISO 11904-1 or GRAS KEMAR).
• Speakers: 130 models measured on a Klippel NFS (ASR, Erin's Audio Corner, via Spinorama at commit acc757bb), including 69 studio monitors (Genelec, Neumann, Kali, Adam, JBL, Kii, Focal, Yamaha…). When Spinorama only has polar data, the CEA-2034 curves are computed locally. The computation matches Klippel's software exactly, which was verified on the KEF R3.
• Room: dimensions, acoustic treatment rate (share of the walls covered with absorbers), bass traps, wall type (masonry / drywall), floor (tile, wood, rug, carpet), ceiling (plaster or acoustic cloud), furniture, rear-wall diffusion, speaker distance from the front wall, ear height. Six presets are included: living room, treated studio, very dead control room, home cinema, bedroom, bare room. Going further: per-wall treatment (front, sides, rear, ceiling), absorber type (5 cm foam, 10 cm wool, 20 cm broadband wool), tweeter height separate from ear height, an off-centre seat (lateral offset), up to 4th-order reflections, rooms up to 30 × 20 × 8 m, listening distance up to 10 m, angle up to 80°, manual RT60 up to 4 s.
• Listening: bass and treble shelves (105 Hz / 10 kHz), left/right balance, preamp.
• Tabs with switches: the settings sit under the room plan in nine tabs (Room, Treatment, Reflections, Reverberation, Placement, Headphones, Listening, Advanced, Output), each with its own on/off switch. A tab that is switched off removes its settings and says what applies instead (room off → free field, placement off → standard stereo triangle, and so on); the values are kept for next time. The Room switch is the only way to remove the room (the room amount goes down to 5 %). The tab bar is always visible: the Settings button in the top bar only collapses the content, and clicking a tab expands it again. No scrolling: everything fits a 1200 × 730 window.
• A/B comparison: the A | B switch in the top bar keeps two complete setups (speakers, room, placement, listening). Switching from one to the other applies it without dropouts; the headphones and the output stay shared.
• Headphones tab: number of filters and preamp of the correction, choice of another measurement of the same model (another source or rig, e.g. oratory1990 or crinacle) and, for an imported raw measurement, a target you can change at any time.
• Seamless changes: Apply starts a second instance (A/B), makes it the default output, waits for the streams to move over, then stops the old one. The output never disappears, so players never pause. In Auto mode every change is applied this way on its own.
• States: start/stop, auto-start at login, and restore of the previous output on stop (even if the service is stopped outside the app).
• Interface: an interactive top-down room plan. Treatment, materials and first-reflection rays are drawn to scale, and you can drag the listener or the speakers. It also shows the simulated frequency response and RT60 meters per octave. The interface is in French or English: switch with the FR | EN button in the top bar (it follows your system language by default). A full command line is also available.

How it works

Five stages are baked into four minimum-phase impulse responses (speaker s → ear e):

IR[s][e] = HP⁻¹ · ( DF⁻¹ · ( LW · Direct[s][e] + ER · Reflections[s][e] ) + SP · Reverb[s][e] )

1. Speaker: the Listening Window filters the direct sound, Early Reflections the first reflections, Sound Power the late reverberation.
2. Room: absorption coefficients per octave band (125 Hz–8 kHz) for every surface. The RT60 is computed per band (Eyring, with air absorption), or imposed manually. A shoebox image-source model (order 0–4) computes the reflections, each one filtered by the materials of every wall it hits. The late reverberation is decorrelated octave-band noise with the RT60 of each band. Its level follows the critical distance, capped below the Schroeder frequency where the statistical model breaks down.
3. Head: every arrival goes through the MIT KEMAR HRTF from its true direction, with exact fractional delays and diffuse-field equalisation.
4. Headphone: the AutoEq parametric EQ becomes IIR, then a minimum-phase impulse response.
5. Output: the IRs are normalised for the worst case (the same signal on both speakers). A separate pipewire -c filter-chain.conf process runs them as a transient systemd user unit.

Usage

GUI

speaker-sim            # or: python3 speaker-sim/main.py

1. 01 Casque (headphones): type a few words (“hd 650 oratory”) and pick your model, or Importer… your own measurement.
2. 02 Enceintes (speakers): filter Studio / Hi-Fi. The tooltip shows the Olive score, the LFX and the measurement origin.
3. Plan: drag yourself (the white circle) to move closer or further, or drag a speaker to change the angle and the distance from the front wall.
4. Settings tabs (under the plan): in Room, pick a preset or set the dimensions and materials; Treatment, Reflections, Reverberation, Placement and Listening refine the rest, and each switch turns its part off. The RT60 meters update live; the light band (0.2–0.4 s) is the usual target for critical listening.
5. Press the power button. With Auto on, later changes apply by themselves, without dropouts.

Command line

speaker-sim --search-headphones "hd 650"
speaker-sim --set-headphone "oratory1990/over-ear/Sennheiser HD 650"
speaker-sim --import-headphone my-measurement.txt --name "My headphones" --target "Harman in-ear 2019"
speaker-sim --list-speakers                    # id, studio/hifi, score
speaker-sim --set-speaker "Neumann KH 150" --start
speaker-sim --status
speaker-sim --stop                             # restores your previous output
speaker-sim --autostart on                     # or off
speaker-sim --export ~/irs                     # portable filters (WAV, true stereo, Equalizer APO)

Windows / macOS (export mode)

Exporter (or --export FOLDER) writes:

• ir_LL/LR/RL/RR.wav and speaker-sim-true-stereo.wav (4 channels: LL, LR, RL, RR);
• speaker-sim.txt for Equalizer APO ≥ 1.2: copy the folder into C:\Program Files\EqualizerAPO\config\ and add Include: speaker-sim.txt to config.txt.

On these systems a pure-Python SOFA reader (h5py) replaces libmysofa.

Files

Everything runs as your user. On Linux the app only writes to:

Location	Content
~/.local/state/devfrp-speaker-sim/config.json	settings
~/.local/state/devfrp-speaker-sim/runtime.json	previous output, active instance
~/.local/state/devfrp-speaker-sim/filters/{a,b}/	IRs + filter-chain.conf per instance
~/.local/state/devfrp-speaker-sim/imports/	imported headphone measurements/EQs
~/.local/state/devfrp-speaker-sim/devfrp-speaker-sim-py-a.service	unit (auto-start)
~/.cache/devfrp-speaker-sim/	AutoEq, Spinorama and KEMAR data (pinned commits)

XDG_STATE_HOME / XDG_CACHE_HOME are honoured. Windows uses %LOCALAPPDATA%\devfrp-speaker-sim; macOS uses ~/Library/Application Support and ~/Library/Caches.

No PipeWire, WirePlumber or Omarchy configuration is modified. The only interaction with WirePlumber is choosing the default output (wpctl set-default), as the volume menu would. Auto-start writes the unit into the state folder and registers it with systemctl --user link + enable. systemd then creates symlinks in ~/.config/systemd/user/ (its only way to know a unit at login); disabling removes them.

Repository

speaker-sim/            Python application (see speaker-sim/README.md)
install.sh              Linux installer (curl | bash)
docs/                   GitHub Pages website
.github/workflows/      CI (tests), Release (.exe / .dmg), Site (Pages)

Publish a version with git tag vX.Y.Z && git push origin vX.Y.Z. GitHub Actions builds the Windows exe and the macOS apps and smoke-tests each binary by exporting real filters. It then creates the release with SHA-256 checksums.

Run the tests with cd speaker-sim && python3 -m unittest discover -s tests -t .. They need no network, PipeWire or systemd (all are mocked).

Troubleshooting

• “libmysofa not found”: install libmysofa, or let the app fall back to h5py (the installer adds it when needed).
• The service stops right away: journalctl --user -u 'devfrp-speaker-sim-py*' -n 50.
• No sound: wpctl status should show Speaker Sim as the default output and devfrp_speaker_sim_py_a.output (or _b) linked to your headphones.
• Too quiet: the filters keep headroom for the worst case. Raise the Préampli (preamp), and watch for clipping in the bass.
• Offline: only data already in the cache can be used.

License

MIT © 2026 devfrp. Data: AutoEq (MIT, Jaakko Pasanen), Spinorama (GPL-3.0, Pierre Aubert) with ASR / Erin's Audio Corner Klippel measurements, MIT KEMAR (Gardner & Martin, MIT Media Lab). The data is downloaded on demand, never redistributed. Fonts: Space Grotesk, Inter, JetBrains Mono (SIL Open Font License, see speaker-sim/packaging/fonts).