This PR adds a 4-point cubic hermite spline interpolation for better resampling.
The implementation is taken from "Polynomial Interpolators for High-Quality Resampling of Oversampled Audio" by Olli Niemitalo (http://yehar.com/blog/wp-content/uploads/2009/08/deip.pdf).
Does not dynamically allocate memory and is only ~2x slower than the InplaceLinearResampler. For comparison, sinc interpolation is ~4.5x slower and requires dynamic memory allocations.
The 2nd, -2nd, and sometimes -1st samples will be slightly off since this interpolator requires 4 points, but we assume a repeated sample when these are not available at the ends rather than adding 2-samples of delay. The 0th sample is always the same as the original audio since its corresponding x value is always 0.
From some experiments in Ableton, the spectrum shows about ~20 dB of improvement for a single sine tone going from 44.1 to 48 kHz at 2048 buffer size. From the frequency response of the hermite interpolator compared to the linear interpolater, we should see ~3 dB improvement in attenuation in the higher frequencies above 10 kHz.
This PR adds a 4-point cubic hermite spline interpolation for better resampling.
The implementation is taken from "Polynomial Interpolators for High-Quality Resampling of Oversampled Audio" by Olli Niemitalo (http://yehar.com/blog/wp-content/uploads/2009/08/deip.pdf). Does not dynamically allocate memory and is only ~2x slower than the
InplaceLinearResampler. For comparison, sinc interpolation is ~4.5x slower and requires dynamic memory allocations.The 2nd, -2nd, and sometimes -1st samples will be slightly off since this interpolator requires 4 points, but we assume a repeated sample when these are not available at the ends rather than adding 2-samples of delay. The 0th sample is always the same as the original audio since its corresponding x value is always 0.
From some experiments in Ableton, the spectrum shows about ~20 dB of improvement for a single sine tone going from 44.1 to 48 kHz at 2048 buffer size. From the frequency response of the hermite interpolator compared to the linear interpolater, we should see ~3 dB improvement in attenuation in the higher frequencies above 10 kHz.
44.1 kHz to 48 kHz (previous linear interpolator)
44.1 kHz to 48 kHz (hermite interpolator)
88.2 kHz to 48 kHz (previous linear interpolator)
88.2 kHz to 48 kHz (hermite interpolator)
cc: @Fyfe93