computeBeatHisto.py

Go to the documentation of this file.

# -*- coding: utf-8 -*-
 
import numpy as np
 
from pyACA.computeSpectrogram import computeSpectrogram
from pyACA.computeNoveltyFunction import computeNoveltyFunction
from pyACA.ToolPreprocAudio import ToolPreprocAudio
from pyACA.ToolComputeHann import ToolComputeHann
from pyACA.ToolReadAudio import ToolReadAudio
 
 
 
def computeBeatHisto(x, f_s, cMethod='FFT', afWindow=None, iBlockLength=1024, iHopLength=8):
    # compute window function for FFT
    if afWindow is None:
        afWindow = ToolComputeHann(iBlockLength)
 
    assert (afWindow.shape[0] == iBlockLength), "parameter error: invalid window dimension"
 
    # pre-processing
    x = ToolPreprocAudio(x)
 
    # novelty function
    [d, t, peaks] = computeNoveltyFunction('Flux', x, f_s, afWindow, iBlockLength, iHopLength)
 
    if cMethod == 'Corr':
        # compute autocorrelation of result
        r_dd = np.correlate(d, d, "full") / np.dot(d, d)
        r_dd = r_dd[np.arange(d.shape[0], r_dd.size)]
 
        Bpm = np.flip(60 / t[np.arange(1, t.shape[0])])
        T = np.flip(r_dd)
 
    elif cMethod == 'FFT':
        # compute the magnitude spectrum of result
        iHistoLength = 65536
        afWindow = np.zeros(2*iHistoLength)
        afWindow[np.arange(0, iHistoLength)] = ToolComputeHann(iHistoLength)
        f_s = f_s / iHopLength
        if len(d) < 2 * iHistoLength:
            d = [d, np.zeros([1, 2 * iHistoLength - len(d)])]
 
        [D, f, t] = computeSpectrogram(d, f_s, afWindow, 2*iHistoLength, iHistoLength/4)
 
        T = D.mean(axis=1, keepdims=True)
 
        # restrict Bpm range
        Bpm = f * 60
        lIdx = np.argwhere(Bpm < 30)[-1]
        hIdx = np.argwhere(Bpm > 200)[0]
        T = T[np.arange(lIdx, hIdx)]
        Bpm = Bpm[np.arange(lIdx, hIdx)]
    else:
        T = 0
        Bpm = 0
 
    return T, Bpm
 
 
def computeBeatHistoCl(cInPath, cOutPath):
    [f_s, afAudioData] = ToolReadAudio(cInPath)
 
    [T, Bpm] = computeBeatHisto(afAudioData, f_s)
 
    result = np.vstack((T, Bpm))
 
    np.savetxt(cOutPath, result)
 
 
if __name__ == "__main__":
    import argparse
 
    parser = argparse.ArgumentParser(description='Compute simple beat histogram of wav file')
    parser.add_argument('--infile', metavar='path', required=False,
                        help='path to input audio file')
    parser.add_argument('--outfile', metavar='path', required=False,
                        help='path to output file')
 
    args = parser.parse_args()
    cInPath = args.infile
    cOutPath = args.outfile
 
    # only for debugging
    if __debug__:
        if not cInPath:
            cInPath = "../../ACA-Plots/audio/sax_example.wav"
        if not cOutPath:
            cOutPath = "c:/temp/out.txt"
 
    # call the function
    computeBeatHistoCl(cInPath, cOutPath)