def testHFCOptimalThreshold(self):
thresholds = map(lambda x: x / 20, range(0, 40, 5))
best_evals = None
best_f = 0
best_threshold = 0
for threshold in thresholds:
evals = []
for f in self.soundfiles:
truthname = self.truth_dir + f.split(".")[0] + ".txt"
srcname = self.sound_dir + f
signal, fs = librosa.load(srcname)
result = onset_detection.detect_onsets(signal,
fs,
rt="hfc",
units="time",
split_bands=True,
threshold=threshold)
truth = mir_eval.io.load_events(truthname)
evals.append(mir_eval.onset.evaluate(truth, result))
f = averageOfMetric(evals, "F-measure")
if f > best_f:
best_f = f
best_threshold = threshold
best_evals = evals
print("Optimal threshold", best_threshold)
printEvaluation("HFC optimal threshold", self.metrics, best_evals)
def testPD(self):
evals = []
for f in self.soundfiles:
truthname = self.truth_dir + f.split(".")[0] + ".txt"
srcname = self.sound_dir + f
signal, fs = librosa.load(srcname)
result = onset_detection.detect_onsets(signal,
fs,
rt="pd",
units="time")
truth = mir_eval.io.load_events(truthname)
evals.append(mir_eval.onset.evaluate(truth, result))
printEvaluation("PD onset detection", self.metrics, evals)
def testHFCMultiband(self):
evals = []
for f in self.soundfiles:
truthname = self.truth_dir + f.split(".")[0] + ".txt"
srcname = self.sound_dir + f
signal, fs = librosa.load(srcname)
result = onset_detection.detect_onsets(signal,
fs,
rt="hfc",
units="time",
split_bands=True)
truth = mir_eval.io.load_events(truthname)
evals.append(mir_eval.onset.evaluate(truth, result))
printEvaluation("HFC onset detection with multiple bands",
self.metrics, evals)
def detect(sound):
waveform = sound[:]
waveform = np.abs(waveform)
#show_waveform(waveform, "Squared wave", 311)
# sw = smooth_wave(waveform)
# show_waveform(sw, "Smoothed wave", 312)
# plt.show()
candidates = onset_detection.detect_onsets(waveform)
print(candidates)
# dis = np.zeros(waveform.shape)
# dis -= 1
# for i in candidates:
# dis[i] = 1
# plt.plot(dis.tolist(), 'g')
return candidates