def predict_sample(self, frames):
freq, sp = clip_to_freq(frames)
smooth_sp = smooth_freq(freq, sp, cutoff=600)
lm_x, lm_y = local_maxima(freq, smooth_sp)
sorted_maxima = [x for (y, x) in sorted(zip(lm_x, lm_y))]
if len(sorted_maxima) < 100:
sorted_maxima += [0 for i in range(100)]
top_100 = sorted_maxima[:100]
top_x = np.array(top_100[:self.sample_size]).flatten()
return self.classifier.predict(top_x)
def build_classifier(dir_name, sample_size=5, c="gnb"):
files = os.listdir(dir_name)
wav_files = [x for x in files if x.endswith(".wav")]
classifier = AudioClassifier(sample_size, c)
for fn in wav_files:
file_full = os.path.join(dir_name, fn)
frames = sample_audio(file_full, start=0)
freq, sp = clip_to_freq(frames)
smooth_sp = smooth_freq(freq, sp, cutoff=600)
lm_x, lm_y = local_maxima(freq, smooth_sp)
sorted_maxima = [x for (y, x) in sorted(zip(lm_x, lm_y))]
if len(sorted_maxima) < 100:
sorted_maxima += [0 for i in range(100)]
top_100 = sorted_maxima[:100]
top_x = np.array(top_100[:classifier.sample_size]).flatten()
class_name = fn.split("_")[0]
refit = False
if len(set(classifier.classifier_y)) > 1:
refit = True
classifier.add_point(top_x, class_name, refit)
return classifier
