def test_sound_recognition_v2():
files = [
"samples/Sonnette/sonnette", "samples/Fire_Alarm/fire_alarm",
"samples/Phone_Ring/phone"
]
dico = np.zeros([513, 0])
print(dico.shape)
for file in files:
stft = np.zeros([513, 0])
for i in range(1, 5):
rate, signal = wav.read(file + str(i) + ".wav")
stft = np.concatenate((stft, spectrum.get_stft(signal / 1.0)),
axis=1)
dico_plus, _ = get_nmf(stft, 3)
dico = np.concatenate((dico, dico_plus), axis=1)
for file in files:
rate2, signal2 = wav.read(file + "5.wav")
stft2 = spectrum.get_stft(signal2 * 1.0)
activations = get_activations(stft2, dico, 3)
plt.clf()
for i in range(0, 9):
plt.subplot(3, 3, i + 1)
plt.title("Ligne " + str(i))
plt.stem(activations[i, :])
plt.show()
def process_audio(self, audio):
"""Compute spectrum from audio source, and call process_spectrum with
the result"""
self.current_audio = np.concatenate((self.current_audio, audio))
spectrum = get_stft(self.current_audio)
if spectrum.size > 0:
self.process_spectrum(spectrum)
# current functions parse the whole audio, so we let nothing in current_audio
# (parsing the whole data regardless of its size, may result in artifacts
# in the reconstructed spectrum)
self.current_audio = np.array([])
def add_dictionary_entry(self, tag, entry):
"""Add a sound to be recognized by Big Alison.
entry: a sound sample containing mostly the sound to recognize."""
stft = get_stft(entry)
dico, _ = nmf.get_nmf(stft, self.components_per_tag)
if self.dictionary is None:
self.dictionary = dico
else:
self.dictionary = np.concatenate((self.dictionary, dico), axis=1)
range_stop = self.dictionary.shape[1]
range_start = range_stop - dico.shape[1]
self.tags[tag] = TagInfo(range(range_start, range_stop))
self._reset_sound_processing()
def demo_nmf():
"""
Show main components from nmf decomposition of an example sound
"""
# D = get_dictionnary(0).components_
# print(D.shape)
rate, signal = wav.read("samples/door-bell01.wav")
stft = spectrum.get_stft(signal[:, 0] / 1.0)
components, activation = dcp.decompose(stft)
i = 0
# first frame spectrum
spectrum = stft[:, i]
# first frame component activation
line = activation[:, i]
# Extract spectrum for the 4 strongest activation on the first frame
ind = np.argsort(line)[-4:]
maincomps = components[:, ind]
# Plot
plt.figure(figsize=(7, 7))
plt.subplot(6, 1, 1)
plt.title("Spectrum")
plt.stem(spectrum)
plt.subplot(6, 1, 2)
plt.title("Components")
plt.stem(line)
for n in range(4):
plt.subplot(6, 1, n + 3)
plt.stem(maincomps[:, n])
plt.show()
waveFile.close()
files = [
"../samples/Sonnette/sonnette", "../samples/Fire_Alarm/fire_alarm",
"../samples/Phone_Ring/phone"
]
dico = np.zeros([513, 0])
print(dico.shape)
for file in files:
stft = np.zeros([513, 0])
for i in range(1, 4):
rate, signal = wav.read(file + str(i) + ".wav")
stft = np.concatenate((stft, spectrum.get_stft(signal / 1.0)), axis=1)
dico_plus, _ = get_nmf(stft, 3)
dico = np.concatenate((dico, dico_plus), axis=1)
for file in files:
rate2, signal2 = wav.read("audio_Record.wav")
stft2 = spectrum.get_stft(signal2 * 1.0)
activations = get_activations(stft2, dico, 3)
plt.clf()
for i in range(0, 9):
plt.subplot(3, 3, i + 1)
plt.title("Ligne " + str(i))
plt.stem(activations[i, :])
def test_sound_recognition():
"""
Example on how to use get_nmf and get_activations
"""
rate, signal = wav.read("samples/Sonnette/sonnette1.wav")
stft = spectrum.get_stft(signal * 1.0)
for file in [
"samples/Sonnette/sonnette", "samples/Fire_Alarm/fire_alarm",
"samples/Phone_Ring/phone"
]:
for i in range(1, 5):
rate, signal = wav.read(file + str(i) + ".wav")
stft = np.concatenate((stft, spectrum.get_stft(signal / 1.0)),
axis=1)
print(stft.shape)
# obtain dictionary with NMF
dico, base_act = get_nmf(stft, 8)
rate2, signal2 = wav.read("samples/Sonnette/sonnette5.wav")
stft2 = spectrum.get_stft(signal2 * 1.0)
# get activations using the previously computed dictionary
# Here it's used on the same sound, but we can use it on different
# sounds in order to classify them
activations = get_activations(stft2, dico, 3)
i = 100
j = 100
frame_act1 = base_act[:, i]
frame_act2 = activations[:, j]
plt.figure(figsize=(7, 7))
plt.subplot(3, 2, 1)
plt.title("Spectrum for sample a")
plt.stem(stft[:, i])
plt.subplot(3, 2, 2)
plt.title("Spectrum for sample b")
plt.stem(stft2[:, j])
plt.subplot(3, 2, 3)
plt.title("Reconstitution of sample b")
plt.stem(np.dot(dico, frame_act2))
plt.subplot(3, 2, 4)
plt.title("Ligne 1")
plt.stem(activations[0, :])
plt.subplot(3, 2, 5)
plt.title("Ligne 2")
plt.stem(activations[1, :])
plt.subplot(3, 2, 6)
plt.title("Ligne 3")
plt.stem(activations[2, :])
plt.show()
import alison.spectrum as spectrum
from alison.nmf import *
from scipy.ndimage.interpolation import shift
files = [
"../samples/Sonnette/sonnette", "../samples/Fire_Alarm/fire_alarm",
"../samples/Phone_Ring/phone"
]
j = 1
#-------- Fire alarm
rate, signal = wav.read("../samples/Fire_Alarm/fire_alarm1.wav")
stft = spectrum.get_stft(signal / 1.0)
dico, _ = get_nmf(stft, 3)
activations = get_activations(stft, dico, 3)
for c in range(0, 3):
plt.subplot(3, 2, j)
plt.title("dico" + "c" + str(c + 1))
plt.stem(dico[:, c])
j = j + 1
plt.subplot(3, 2, j)
plt.title("acti" + "l" + str(c + 1))
plt.stem(activations[c, :])
j = j + 1
plt.suptitle("Fire alarm 1")
plt.show()
# -------- Complex sounds