DSP

This library does not provide complete DSP paths to render ADM content, but does contain some components which can be used to do so. DSP components are defined in Namespace ear::dsp.

FFT interface

BlockConvolver objects use a user-provided FFT implementation. These are provided by implementing FFTImpl<float> (and therefore FFTPlan<float> and FFTWorkBuf) for the FFT library you wish to use, and passing an instance to BlockConvolver::Context::Context().

An implementation for KISS FFT is provided by default, and may be obtained by calling get_fft_kiss(). The implementation of this (in src/fft_kiss.cpp) may be a useful example to show how the FFT interface should be implemented.

Rendering DirectSpeakers

The gains calculated for DirectSpeakers channels using the GainCalculatorDirectSpeakers should be applied directly to the input audio channel to produce the output audio channels. DirectSpeakers metadata should not be dynamic (there should be a single audioBlockFormat in each audioChannelFormat), so gains should not be interpolated inside blocks, though should be interpolated if metadata is changed by the user.

This may be applied using the GainInterpolator with LinearInterpVector.

Rendering Objects

The audio processing for Objects content is defined in [bs2127] section 7.1. The structure used is as in Fig. 1.

Fig. 1 Signal processing for m Objects with n output channels

This can be built from the following components:

• GainInterpolator with LinearInterpVector to interpolate and apply the gains to the incoming audio (the interp blocks in Fig. 1).

• DelayBuffer<float> with decorrelatorCompensationDelay() samples delay to compensate for the decorrelator delays.

• BlockConvolver objects with filters calculated using designDecorrelators() to decorrelate the signals.

• VariableBlockSizeAdapter<float> to allow the use of BlockConvolver objects with variable-size sample blocks. This could be used to wrap just the BlockConvolver::process() calls, or the whole processing chain (recommended). If only the block convolvers are adapted, then the compensation delay will need to be increased by VariableBlockSizeAdapter::get_delay() samples.

Rendering HOA

As with DirectSpeakers, HOA metadata should not be dynamic, so the calculated matrices can be applied directly to the input audio.

The decode matrices calculated for HOA channels using the GainCalculatorHOA should be applied directly to the input audio channels to produce the output audio channels. As with DirectSpeakers, HOA metadata should not be dynamic (there should be a single audioBlockFormat in each audioChannelFormat), so gains should not be interpolated inside blocks, though should be interpolated if metadata is changed by the user.

This may be applied using the GainInterpolator with LinearInterpMatrix.