I've added sonus.pipeline.fft (and left legacy sonus.pipeline.fft_df) which should implement FFT for the windowed data. The function sonus.pipeline.fft appends columns to the X data, so we'll need to be careful keeping track of which column is which moving forwards. The fft also now has a cutoff so that all the frequency data is in the range of human speech (this is customizable).
I've added
sonus.pipeline.fft(and left legacysonus.pipeline.fft_df) which should implement FFT for the windowed data. The functionsonus.pipeline.fftappends columns to the X data, so we'll need to be careful keeping track of which column is which moving forwards. The fft also now has a cutoff so that all the frequency data is in the range of human speech (this is customizable).Here's some example code and plots:
import sonus.pipelineimport sonus.parseimport sonus.examplesdata = pickle.load(open('data/vox2_test_aac_wav.pkl', 'rb'))data = data.iloc[0:100]X_train, X_test, y_train, y_test = sonus.parse.window_data(data['audio'], data['id'], 0.5, 0.5, 16000, 0.25)fft_x, fftout = sonus.pipeline.fft(X_train, return_x=True)sonus.examples.plot_fft(fftout, fft_x, 50)sonus.examples.plot_fft(fftout, fft_x, 100)