yoyololicon / kazane

Simple sinc interpolation in PyTorch.

Kazane: simple sinc interpolation for 1D signal in PyTorch

Kazane utilize FFT based convolution to provide fast sinc interpolation for 1D signal when your sample rate only needs to change by an integer amounts; If you need to change by a fraction amounts, checkout julius.

Installation

pip install kazane

or

pip install git+https://github.com/yoyololicon/kazane

for latest version.

Usage

import kazane
import torch

signal = torch.randn(8, 2, 44100)

# downsample by an amount of 3
decimater = kazane.Decimate(3)
resampled_signal = decimater(signal)

# upsample by an amount of 2
upsampler = kazane.Upsample(2)
resampled_signal = upsampler(signal)

# you can also control number of zeros, roll-off frequency of the sinc interpolation kernel
decimater = kazane.Decimate(3, num_zeros=24, roll_off=0.9)

# use other types of window function for the sinc kernel
upsampler = kazane.Upsample(2, window_func=torch.blackman_window)

Benchmarks on CUDA

Using the benchmark scripts at bench, you can see that FFT can gives some speed improvements when the sample rate changes with some common integer numbers.

[---------- Down sample ----------]
               |  julius  |  kazane
2 threads: ------------------------
      rate: 2  |   52.2   |   52.4 
      rate: 3  |   66.5   |   36.1 
      rate: 5  |   94.8   |   30.0 
      rate: 7  |  121.7   |   42.3 

Times are in milliseconds (ms).

[----------- Up sample -----------]
               |  julius  |  kazane
2 threads: ------------------------
      rate: 2  |   48.8   |   39.0 
      rate: 3  |   68.1   |   51.6 
      rate: 5  |  112.5   |   78.9 
      rate: 7  |  159.4   |  108.0 

Times are in milliseconds (ms).