ref: f98063b27bf9014142bfaae0e7a63a1d5cd49cae
parent: f131ba849aaae809214f5c74c97a69981c361692
author: Paul Brossier <piem@piem.org>
date: Wed Feb 1 21:39:05 EST 2017
python/demos/demo_bench_yin.py: add yin/yinfft benchmark on sine waves
--- /dev/null
+++ b/python/demos/demo_bench_yin.py
@@ -1,0 +1,50 @@
+#! /usr/bin/env python
+
+import numpy as np
+from aubio import pitch
+import pylab as plt
+
+buf_size = 2048 * 1
+hop_size = buf_size // 4
+
+samplerate = 44100
+minfreq = 40
+maxfreq = 6000
+
+def sinewave(freq, duration, samplerate = samplerate):
+ """ generate a sinewave """
+ length = hop_size
+ while length < duration * samplerate:
+ length += hop_size
+ return np.sin( 2. * np.pi * np.arange(length) * freq / samplerate ).astype("float32")+
+def get_stats_for_pitch_method(method, freqs, samplerate = samplerate):
+ """ for a given pitch method and a list of frequency, generate a sinewave
+ and get mean deviation """
+ means = np.zeros(len(freqs))
+ medians = np.zeros(len(freqs))
+ for freq, fn in zip(freqs, range(len(freqs))):
+ s = sinewave(freq, .50).reshape(-1, hop_size)
+ #s = (sinewave(freq, .50) + .0*sinewave(freq/2., .50)).reshape(-1, hop_size)
+ p = pitch(method, buf_size, hop_size, samplerate = samplerate)
+ candidates = np.zeros(len(s))
+ #samples = np.zeros(buf_size)
+ for frame, i in zip(s, range(len(s))):
+ candidates[i] = p(frame)[0]
+ # skip first few candidates
+ candidates = candidates[4:]
+ means[fn] = np.mean(candidates[candidates != 0] - freq)
+ medians[fn] = np.median(candidates[candidates != 0] - freq)
+ print (freq, means[fn], medians[fn])
+ return means, medians
+
+if __name__ == '__main__':
+ freqs = np.arange(minfreq, maxfreq, 1.)
+ modes = ["yin", "yinfft"]
+ for mode in modes:
+ means, medians = get_stats_for_pitch_method(mode, freqs)
+ plt.figure()
+ plt.plot(freqs, means, 'g-')
+ plt.plot(freqs, medians, 'r--')
+ #plt.savefig(mode + '_deviations_test.png', dpi=300)
+ plt.show()