IRCAM Multichannel to Binaural — User Guide

What this does

IRCAM Multichannel to Binaural converts a loudspeaker-oriented multichannel Sound into a two-channel headphone rendering. Praat exports a temporary multichannel WAV; the Python bridge calls spat5.virtualspeakers~; Spat5 applies the selected loudspeaker layout, HRTF, ITD scale, and optional room model; the resulting binaural WAV is analysed and returned to Praat.

Quick start

1. Install IRCAM Spat5 and locate the folder containing spat5.virtualspeakers~ (or spat5.virtualspeakers~.exe on Windows).
2. Place spat_binaural_bridge.py in the Praat AudioTools script folder.
3. Select exactly one multichannel Sound in Praat.
4. Run IRCAM_Multichannel_to_Binaural.praat.
5. Choose Auto-detect from channel count unless you have a specific reason to select a layout manually.
6. Choose an HRTF preset and ITD value.
7. Leave Headroom on Auto safe render for normal use.
8. Optionally mute the LFE channel, normalize the accepted binaural render to the requested output peak, draw the visualization, or play the result.

Signal flow & safe render

Each binaural ear is a sum of several HRIR-filtered input channels. Even when no individual input channel clips, the summed binaural result can exceed full scale. The tool therefore does not rely on a single fixed pre-gain.

In Auto safe render, the first attenuation is estimated from source peak, number of active channels, and additional HRTF headroom. Every rendered output is then measured. If PCM clipping is detected, that render is rejected and Spat5 is called again from the original source with another 6 dB of attenuation, up to five attempts.

Supported layouts & channel order

Auto-detect maps the Sound's channel count to one of these layouts. If a layout is chosen manually, the script verifies that its required channel count matches the Sound before rendering.

HRTF, ITD & room

KEMAR presets

Custom SOFA

Choose SOFA custom / none to supply a SOFA dataset name through Sofa_file. In this mode the separate Room preset becomes active and may be set to none, hall, livingroom, or studio.

ITD percent

ITD_percent scales the interaural time-difference component used by the Spat5 render. 100% is the normal reference value; lower values reduce temporal left/right separation and higher values exaggerate it.

Headroom & output level

Auto safe render

This is the recommended mode. The tool estimates a conservative initial input attenuation, renders the binaural signal, measures the result, and automatically tries again with an additional 6 dB attenuation if clipping is found.

Manual pre-gain

Manual mode performs one render using Manual_pre_gain_dB. The gain is still limited when necessary so that the temporary input WAV itself does not exceed approximately −1 dBFS. The rendered binaural file is still analysed for clipping.

Normalise output

When Normalise_output is enabled, a clean accepted render is scaled so that its final peak reaches Output_peak_dBFS. This happens only after the Spat5 render has passed the clipping check.

If Spat5 returns a floating-point WAV whose samples exceed ±1, the values have not been hard-clipped. The script preserves the render and scales it into a safe playback range.

LFE handling

Mute_LFE is available for layouts whose LFE position is known in the script: 5.1, 7.1, 7.1.2, and 7.1.4. In these layouts channel 4 is treated as LFE and can be set to zero before binaural rendering.

The option is ignored for layouts that do not have a known LFE channel.

Channel-order test Sound

Enable Create_channel_order_test_Sound to create a diagnostic multichannel Sound instead of rendering the selected source.

The generated test contains one approximately 0.8-second pink-noise burst per channel, one channel at a time, at about −20 dBFS peak. Burst k starts near second k−1.

1. Choose the layout you want to verify.
2. Create the test Sound.
3. Run the binaural tool again on that test Sound using KEMAR / neutral and Auto safe render.
4. Listen to the bursts in sequence and confirm that each appears from the expected loudspeaker direction.

Controls

Visualization

The Picture-window figure shows both the audio result and the level-management process.

• A — Input: channel 1 and, when present, channel 2 at their true source level.
• B — Binaural output: left and right ear waveforms, with ±1 full-scale guides and final peak.
• C — Safe render: one column per render attempt; clipped attempts are distinguished from the accepted clean render, together with the input attenuation used.
• D — Level ledger: source peak, accepted input attenuation, render peak, normalization gain, output peak, and net source-to-output gain.
• Summary strip: layout, assumed channel order, HRTF, ITD, room, headroom mode, output format, duration, and descriptive L/R correlation.

Render diagnostics

The Python bridge analyses the binaural WAV after each render. It reports:

• output format, sample rate, frame count, and stereo channel count;
• peak and RMS level for each ear;
• crest factor;
• number of full-scale PCM samples;
• longest consecutive full-scale run;
• for floating-point output, samples whose absolute value exceeds 1;
• L/R zero-lag correlation.

Output behavior

A successful render creates a stereo Sound named:

Temporary WAV, log, and statistics files are deleted after a clean accepted render. If the external render fails, the log is preserved and printed in the Info window. If repeated clipping prevents acceptance, the final clipped render may be loaded for inspection but is explicitly reported as unaccepted.

The Info window also prints a level ledger containing the source peak, accepted input attenuation, number of render attempts, Spat5 render peak, normalization gain, final output peak, net gain, and L/R correlation.

Troubleshooting

Spat5 tool not found

Set Tools_folder to the directory that actually contains spat5.virtualspeakers~ or spat5.virtualspeakers~.exe.

Python not found

The script probes common Python commands and requires a working Python 3 installation on the system PATH. NumPy is optional; when available it speeds up output analysis.

Layout rejected

The selected layout must match the Sound's channel count. Auto-detect supports 1, 2, 4, 5, 6, 7, 8, 10, 12, and 24 channels.

Spatial positions sound wrong

The likely cause is channel-order mismatch. Generate and render the channel-order test Sound to verify the layout one channel at a time.

Render clips

Use Auto safe render. The tool will progressively lower the input and repeat the Spat5 render. A clipped PCM render is never accepted as clean.

Output is quieter or louder than expected

Inspect the Level ledger. The final level depends on the attenuation required for a clean render and on the optional output normalization target.

L/R correlation is high

This is not automatically an error. Frontal or strongly correlated material can produce high binaural correlation. Judge the spatial result together with the source layout and listening position.

Dependencies

• Praat
• IRCAM Spat5 command-line tools, including spat5.virtualspeakers~
• Python 3
• NumPy is optional and used only to accelerate file analysis when available