niswegmann / Resampler

Small header-only C++ library for real-time sample rate conversion with a fractional ratio.

Resampler

Resampler is a small header-only C++ library for real-time sample rate conversion with a fractional ratio.

Here's and example of how to use it. Say we want to convert from 44.1 Hz to 48 Hz in stereo.

Start by creating a Resampler instance with 2 input and 2 output channels and store it somewhere.

double const source_sample_rate = 44100, target_sample_rate = 48000;

Resampler<2, 2> resampler (source_sample_rate, target_sample_rate);

Allocate your target buffers (for processing the signal at 48 kHz):

float * const target_input_buffers [2] =
{
    malloc(max_target_sample_count * sizeof(float)),
    malloc(max_target_sample_count * sizeof(float))
};

float * const target_output_buffers [2] =
{
    malloc(max_target_sample_count * sizeof(float)),
    malloc(max_target_sample_count * sizeof(float))
};

Make sure that max_target_sample_count is greater than or equal to ceil(max_source_sample_count * target_sample_rate / source_sample_rate), where max_source_sample_count is the maximum sample count for the source buffers (that contains the provided signal at 44.1 kHz).

In your process callback, use the resampler to process the 44.1 kHz signal at 48 kHz instead.

...

resampler.process
(
    source_input_buffers,
    source_output_buffers,
    target_input_buffers,
    target_output_buffers,
    sample_count,
    [](float * const input_buffers [2], float * const output_buffers [2], int sample_count)
    {
        ... // Do your DSP on the resampled buffers.
    }
);

...

Resampler also works the other way around, if you want to process a 48 kHz signal at 44.1 kHz.

Internally, Resampler uses a 3rd-order polynomial (cubic hermite) interpolator on a 4-point window. A slightly more complicated version of linear interpolation. Quality-wise, Resampler works best if you oversample the source signal first by a factor of 4, 8 or 16, depending on your needs.