Algorithms
Here are explanations about the different algorithms Impulcifer has for processing the sine sweep recordings into HRIRs.
Headphones and measurement microphones have a frequency responses of their own and these will have to be compensated in order to have real HRIR and virtualized HRIR sound the same. Frequency response of headphones and measurement microphones can be compensated by measuring frequency response of headphones with the measurement microphones in ears and equalizing the measured frequency response flat. This way the frequency response at the ear canal opening will be the same when listening to the speakers or virtualizing the speakers on headphones.
Impulcifer can do this headphone compensation as long as the data directory contains exponential sine sweep measurement
file called headphones.wav. Impulcifer will calculate the frequency response of headphones-microphones system from the
sine sweep recording and creates a equalization FIR filters by inverting the frequency responses of left and right ear.
All left ear impulse responses of the measured HRIR are then equalized with the left side equalization FIR filter and
right ear impulse responses with the right side equalization FIR filter. Impulcifer will take the absolute gain values
of headphone-microphone frequency responses into account so any channel imbalances will be compensated as well.
Because the headphone compensation recording is done with the same binaural measurement micrphones the equalization filters will equalize the microphone frequency responses too. This creates the benefit of not needing to have microphones with flat frequency response.
Script compensation.py and recordings at data/compensation serve as a proof that headphone compensation works as
intended. data/compensation contains three recordings: normal HRIR recording, heaphone sine sweep recording and
virtualized HRIR recording. Virtualized HRIR is done on headphones with virtualization processing using HRIR created
with the first two recordings. Graphs below show that the frequency responses of raw HRIR and virtualized HRIR match
each other very closely. Narrow notches in the error curve are caused by smart smoothing which avoids narrow positive
gain spikes in the equalization filters.
This compensation scheme is based on a naive assumption that if a frequency response is the same at the ear canal opening when listening to speakers versus when listening virtualized speakers on headphones then the frequency responses will be the same at the ear drum. In truth this assumption does not hold true for most headphones. Even large around ear headphones will affect ear canal resonances changing the frequency response at the ear drum. Unfortunately it is not possible to measure headphone's effect of ear canal resonances with binaural microphones that sit at the ear canal entrance.
- Overview
- Impulse response estimation
- Pre-ringing cancellation
- Tracking filter
- Channel delay syncing
- Channel delay calculation (equal distance to point between ears)
- Reverberation time adjustment
- Early reflection management
- Room correction
- Plots