python/demos/demo_pitch.py

   1 #! /usr/bin/env python
   2
   3 import sys
   4 from aubio import source, pitch
   5
   6 if len(sys.argv) < 2:
   7     print("Usage: %s <filename> [samplerate]" % sys.argv[0])
   8     sys.exit(1)
   9
  10 filename = sys.argv[1]
  11
  12 downsample = 1
  13 samplerate = 44100 // downsample
  14 if len( sys.argv ) > 2: samplerate = int(sys.argv[2])
  15
  16 win_s = 4096 // downsample # fft size
  17 hop_s = 512  // downsample # hop size
  18
  19 s = source(filename, samplerate, hop_s)
  20 samplerate = s.samplerate
  21
  22 tolerance = 0.8
  23
  24 pitch_o = pitch("yin", win_s, hop_s, samplerate)
  25 pitch_o.set_unit("midi")
  26 pitch_o.set_tolerance(tolerance)
  27
  28 pitches = []
  29 confidences = []
  30
  31 # total number of frames read
  32 total_frames = 0
  33 while True:
  34     samples, read = s()
  35     pitch = pitch_o(samples)[0]
  36     #pitch = int(round(pitch))
  37     confidence = pitch_o.get_confidence()
  38     #if confidence < 0.8: pitch = 0.
  39     print("%f %f %f" % (total_frames / float(samplerate), pitch, confidence))
  40     pitches += [pitch]
  41     confidences += [confidence]
  42     total_frames += read
  43     if read < hop_s: break
  44
  45 if 0: sys.exit(0)
  46
  47 #print pitches
  48 import os.path
  49 from numpy import array, ma
  50 import matplotlib.pyplot as plt
  51 from demo_waveform_plot import get_waveform_plot, set_xlabels_sample2time
  52
  53 skip = 1
  54
  55 pitches = array(pitches[skip:])
  56 confidences = array(confidences[skip:])
  57 times = [t * hop_s for t in range(len(pitches))]
  58
  59 fig = plt.figure()
  60
  61 ax1 = fig.add_subplot(311)
  62 ax1 = get_waveform_plot(filename, samplerate = samplerate, block_size = hop_s, ax = ax1)
  63 plt.setp(ax1.get_xticklabels(), visible = False)
  64 ax1.set_xlabel('')
  65
  66 def array_from_text_file(filename, dtype = 'float'):
  67     filename = os.path.join(os.path.dirname(__file__), filename)
  68     return array([line.split() for line in open(filename).readlines()],
  69         dtype = dtype)
  70
  71 ax2 = fig.add_subplot(312, sharex = ax1)
  72 ground_truth = os.path.splitext(filename)[0] + '.f0.Corrected'
  73 if os.path.isfile(ground_truth):
  74     ground_truth = array_from_text_file(ground_truth)
  75     true_freqs = ground_truth[:,2]
  76     true_freqs = ma.masked_where(true_freqs < 2, true_freqs)
  77     true_times = float(samplerate) * ground_truth[:,0]
  78     ax2.plot(true_times, true_freqs, 'r')
  79     ax2.axis( ymin = 0.9 * true_freqs.min(), ymax = 1.1 * true_freqs.max() )
  80 # plot raw pitches
  81 ax2.plot(times, pitches, '.g')
  82 # plot cleaned up pitches
  83 cleaned_pitches = pitches
  84 #cleaned_pitches = ma.masked_where(cleaned_pitches < 0, cleaned_pitches)
  85 #cleaned_pitches = ma.masked_where(cleaned_pitches > 120, cleaned_pitches)
  86 cleaned_pitches = ma.masked_where(confidences < tolerance, cleaned_pitches)
  87 ax2.plot(times, cleaned_pitches, '.-')
  88 #ax2.axis( ymin = 0.9 * cleaned_pitches.min(), ymax = 1.1 * cleaned_pitches.max() )
  89 #ax2.axis( ymin = 55, ymax = 70 )
  90 plt.setp(ax2.get_xticklabels(), visible = False)
  91 ax2.set_ylabel('f0 (midi)')
  92
  93 # plot confidence
  94 ax3 = fig.add_subplot(313, sharex = ax1)
  95 # plot the confidence
  96 ax3.plot(times, confidences)
  97 # draw a line at tolerance
  98 ax3.plot(times, [tolerance]*len(confidences))
  99 ax3.axis( xmin = times[0], xmax = times[-1])
 100 ax3.set_ylabel('condidence')
 101 set_xlabels_sample2time(ax3, times[-1], samplerate)
 102 plt.show()
 103 #plt.savefig(os.path.basename(filename) + '.svg')