From Inter res et verba · · 1m 50s
Reverse Engineering Spatial Audio for Linux
A developer reverse-engineered a commercial spatial audio effect using Linux audio tools to create a better listening experience.
In brief
A developer reverse-engineered a commercial spatial audio effect using Linux audio tools to create a better listening experience. A developer created a Linux audio tool that mimics spatial audio effects by analyzing a commercial program's output. Originally reported by Inter res et verba.
First attempt
I began writing a spatial audio program for Linux using PipeWire and an LLM to experiment with audio filters.
“I had no prior audio experience, so I began by using an LLM as a sounding board while experimenting with standard audio filters and psychoacoustic techniques.”
HRTF research
I tried MIT KEMAR database first but found it sounded cheap and inwards; later switched to SADIE II for better room response.
“I tried the mit kemar database first and thought it sounded awful, because it had been recorded in an anechoic chamber, so there were no room echoes, which were a desirable psychoacoustic effect here.”
LLM limitations
The LLM described complex audio terms like perceptual envelopment but provided no actionable guidance for improvement.
“The llm began talking about things like "perceptual envelopment", "late-field decorrelation", "spectral diffusion" and "diffuse spatial energy".”
Re-attempt strategy
I stopped trying to reverse-engineer the program's internal parameters and instead measured its output behavior directly.
“Instead of trying to discover the exact parameters or techniques and algorithms the program used internally, i could study the behaviour of the filter itself.”
Impulse response
I measured the program's response to single impulses to capture how sound decays over time and phase delays.
“If i pass a file containing a single nonzero sample through the filter, everything which appears after it in the recording would've been added by the filter.”
Sweep measurement
I used logarithmic audio sweeps to measure frequency responses while avoiding noise interference from the recording process.
“A shortened logarithmic sweep. Download the audio example”
Steady tone test
I tested how the program's gain changed with amplitude and how long it took to adjust to new sounds.
“I added steady tones at different frequencies and amplitudes, and longer tones going from quiet to loud and back, to see whether the gain changed and how long it took to do so.”
Multitone test
I checked if the program generated new frequencies by playing a multitone signal across eight frequencies.
“The eight-frequency multitone. Download the audio example”
Time-invariance test
I verified the program's response remained consistent over time by repeating the multitone signal three times.
“Three identical copies of the multitone. Download the audio example”
Noise test
I tested if the extracted filter could reproduce its original output using noise samples across frequency ranges.
“Noise covering the low, middle and high frequency ranges in sequence. Download the audio example”
Recovering the filter
I calculated the offset between the recording and original clicks to align the audio responses correctly.
“I had put evenly spaced clicks near the start, which i initially tried finding by correlating the recording with the original clicks.”
Final output
The extracted filter recreated the spatial audio effect with minimal error, showing promise for real-time linux audio enhancement.
“For the main surround setting, the scaled impulses were nearly identical, the combined inputs matched the sum of their separate responses, and the multitone produced almost no new frequencies.”
The big picture
A developer created a Linux audio tool that mimics spatial audio effects by analyzing a commercial program's output.
