def main():
dest = "tests/test_1_note_Csharp3.wav"
tone = librosa.tone(138.59, sr=22050, length=44100)
soundfile.write(dest, tone, 22050)
print("Created {0} with note C#3".format(dest))
dest = "tests/test_1_note_E4.wav"
tone = librosa.tone(329.63, sr=22050, length=44100)
soundfile.write(dest, tone, 22050)
print("Created {0} with note E4".format(dest))
dest = "tests/test_2_notes_E2_F3.wav"
tone = numpy.zeros(44100)
tone += librosa.tone(82.41, sr=22050, length=44100)
tone += librosa.tone(174.61, sr=22050, length=44100)
soundfile.write(dest, tone, 22050)
print("Created {0} with notes E2, F3".format(dest))
dest = "tests/test_2_notes_G3_Asharp4.wav"
tone = numpy.zeros(44100)
tone += librosa.tone(196, sr=22050, length=44100)
tone += librosa.tone(466.16, sr=22050, length=44100)
soundfile.write(dest, tone, 22050)
print("Created {0} with notes G3, A#4".format(dest))
dest = "tests/test_3_notes_G2_B2_G#3.wav"
tone = numpy.zeros(44100)
tone += librosa.tone(98, sr=22050, length=44100)
tone += librosa.tone(123.47, sr=22050, length=44100)
tone += librosa.tone(207.65, sr=22050, length=44100)
soundfile.write(dest, tone, 22050)
print("Created {0} with notes G2, B2, G#3".format(dest))
return 0
def artificial_inst(f: float, sr: float, length: int, phis: dict = None):
"""
Create artificial instrumental sound
Args:
f: fundamental frequency of the instrument
sr: sampling rate
length: sample length of the instrumental sound
phis: phase information for all the harmonics in the artificial
instrument. key: value -> harmonic number:
Returns:
"""
H = (1, 3, 5)
if phis is None:
phis = {h: -np.pi / 2 for h in H}
# add 2, 3, 4, 8 - th subharmonics to enrich sound
piano_sound = 0
for h in H:
piano_sound += 1 / h * librosa.tone(
h * f, sr=sr, length=length, phi=phis[h])
# normalize value
piano_sound /= sum([1 / h for h in H])
def get_phi(f, phi):
return 2 * np.pi * f * length / sr + phi
phis = {h: get_phi(h * f, phis[h]) for h in H}
# put some decay
return piano_sound, phis
def test_mfcc_to_mel(n_mfcc, n_mels, dct_type, lifter):
y = librosa.tone(440.0, sr=22050, duration=1)
mfcc = librosa.feature.mfcc(y=y,
sr=22050,
n_mels=n_mels,
n_mfcc=n_mfcc,
dct_type=dct_type)
# check lifter parameter error
if lifter < 0:
with pytest.raises(librosa.ParameterError):
librosa.feature.inverse.mfcc_to_mel(mfcc * 10**3,
n_mels=n_mels,
dct_type=dct_type,
lifter=lifter)
# check no lifter computations
elif lifter == 0:
melspec = librosa.feature.melspectrogram(y=y, sr=22050, n_mels=n_mels)
mel_recover = librosa.feature.inverse.mfcc_to_mel(mfcc,
n_mels=n_mels,
dct_type=dct_type)
# Quick shape check
assert melspec.shape == mel_recover.shape
# Check non-negativity
assert np.all(mel_recover >= 0)
# check that runtime warnings are triggered when appropriate
elif lifter == 2:
with pytest.warns(UserWarning):
librosa.feature.inverse.mfcc_to_mel(mfcc * 10**3,
n_mels=n_mels,
dct_type=dct_type,
lifter=lifter)
# check if mfcc_to_mel works correctly with lifter
else:
ones = np.ones(mfcc.shape, dtype=mfcc.dtype)
n_mfcc = mfcc.shape[0]
idx = np.arange(1, 1 + n_mfcc, dtype=mfcc.dtype)
lifter_sine = 1 + lifter * 0.5 * np.sin(
np.pi * idx / lifter)[:, np.newaxis]
# compute the recovered mel
mel_recover = librosa.feature.inverse.mfcc_to_mel(ones * lifter_sine,
n_mels=n_mels,
dct_type=dct_type,
lifter=lifter)
# compute the expected mel
mel_expected = librosa.feature.inverse.mfcc_to_mel(ones,
n_mels=n_mels,
dct_type=dct_type,
lifter=0)
# assert equality of expected and recovered mels
np.testing.assert_almost_equal(mel_recover, mel_expected, 3)
def create_tone(freq, temp_dir):
temp_path = temp_dir + '/temp.wav'
tone = librosa.tone(freq / 2, duration=10)
librosa.output.write_wav(temp_path, tone, 44100)
out_path = temp_dir + '/out.wav'
subprocess.call([
'ffmpeg', '-i', temp_path, '-acodec', 'pcm_s16le', '-ac', '1', '-ar',
'44100', out_path
])
return out_path
def __test(frequency, sr, length, duration, phi):
y = librosa.tone(frequency=frequency,
sr=sr,
length=length,
duration=duration,
phi=phi)
if length is not None:
assert len(y) == length
else:
assert len(y) == np.ceil(duration * sr)
def test_mel_to_audio():
y = librosa.tone(440.0, sr=22050, duration=1)
M = librosa.feature.melspectrogram(y=y, sr=22050)
y_inv = librosa.feature.inverse.mel_to_audio(M, sr=22050, length=len(y))
# Sanity check the length
assert len(y) == len(y_inv)
# And that it's valid audio
assert librosa.util.valid_audio(y_inv)
def test_mel_to_audio():
y = librosa.tone(440.0, sr=22050, duration=1)
M = librosa.feature.melspectrogram(y=y, sr=22050)
y_inv = librosa.feature.inverse.mel_to_audio(M, sr=22050, length=len(y))
# Sanity check the length
assert len(y) == len(y_inv)
# And that it's valid audio
assert librosa.util.valid_audio(y_inv)
def __test(frequency, sr, length, duration, phi):
y = librosa.tone(frequency=frequency,
sr=sr,
length=length,
duration=duration,
phi=phi)
if length is not None:
assert len(y) == length
else:
assert len(y) == np.ceil(duration * sr)
def test_chroma_issue1295(freq):
tone_1 = librosa.tone(frequency=freq, sr=22050, duration=1)
chroma_1 = librosa.feature.chroma_stft(y=tone_1,
sr=22050,
n_chroma=120,
base_c=True)
actual_argmax = np.unravel_index(chroma_1.argmax(), chroma_1.shape)
if freq == 261.63:
assert actual_argmax == (118, 0)
elif freq == 440:
assert actual_argmax == (90, 0)
def test_mfcc_to_audio(n_mfcc, n_mels, dct_type):
y = librosa.tone(440.0, sr=22050, duration=1)
mfcc = librosa.feature.mfcc(y=y, sr=22050,
n_mels=n_mels, n_mfcc=n_mfcc, dct_type=dct_type)
y_inv = librosa.feature.inverse.mfcc_to_audio(mfcc, n_mels=n_mels,
dct_type=dct_type,
length=len(y))
# Sanity check the length
assert len(y) == len(y_inv)
# And that it's valid audio
assert librosa.util.valid_audio(y_inv)
def test_mfcc_to_audio(n_mfcc, n_mels, dct_type):
y = librosa.tone(440.0, sr=22050, duration=1)
mfcc = librosa.feature.mfcc(y=y, sr=22050,
n_mels=n_mels, n_mfcc=n_mfcc, dct_type=dct_type)
y_inv = librosa.feature.inverse.mfcc_to_audio(mfcc, n_mels=n_mels,
dct_type=dct_type,
length=len(y))
# Sanity check the length
assert len(y) == len(y_inv)
# And that it's valid audio
assert librosa.util.valid_audio(y_inv)
def test_mfcc_to_mel(n_mfcc, n_mels, dct_type):
y = librosa.tone(440.0, sr=22050, duration=1)
mfcc = librosa.feature.mfcc(y=y, sr=22050,
n_mels=n_mels, n_mfcc=n_mfcc, dct_type=dct_type)
melspec = librosa.feature.melspectrogram(y=y, sr=22050, n_mels=n_mels)
mel_recover = librosa.feature.inverse.mfcc_to_mel(mfcc, n_mels=n_mels,
dct_type=dct_type)
# Quick shape check
assert melspec.shape == mel_recover.shape
# Check non-negativity
assert np.all(mel_recover >= 0)
def test_mfcc_to_mel(n_mfcc, n_mels, dct_type):
y = librosa.tone(440.0, sr=22050, duration=1)
mfcc = librosa.feature.mfcc(y=y, sr=22050,
n_mels=n_mels, n_mfcc=n_mfcc, dct_type=dct_type)
melspec = librosa.feature.melspectrogram(y=y, sr=22050, n_mels=n_mels)
mel_recover = librosa.feature.inverse.mfcc_to_mel(mfcc, n_mels=n_mels,
dct_type=dct_type)
# Quick shape check
assert melspec.shape == mel_recover.shape
# Check non-negativity
assert np.all(mel_recover >= 0)
M = librosa.feature.melspectrogram(y=y, sr=22050)
y_inv = librosa.feature.inverse.mel_to_audio(M, sr=22050, length=len(y))
# Sanity check the length
assert len(y) == len(y_inv)
# And that it's valid audio
assert librosa.util.valid_audio(y_inv)
@pytest.mark.parametrize("n_mfcc", [13, 20])
@pytest.mark.parametrize("n_mels", [64, 128])
@pytest.mark.parametrize("dct_type", [2, 3])
@pytest.mark.parametrize("lifter", [-1, 0, 1, 2, 3])
@pytest.mark.parametrize("y", [librosa.tone(440.0, sr=22050, duration=1)])
def test_mfcc_to_mel(y, n_mfcc, n_mels, dct_type, lifter):
mfcc = librosa.feature.mfcc(y=y,
sr=22050,
n_mels=n_mels,
n_mfcc=n_mfcc,
dct_type=dct_type)
# check lifter parameter error
if lifter < 0:
with pytest.raises(librosa.ParameterError):
librosa.feature.inverse.mfcc_to_mel(mfcc * 10**3,
n_mels=n_mels,
dct_type=dct_type,
lifter=lifter)
def noisy_sinewave(freq, sample_rate, duration):
signal = librosa.tone(freq, sr=sample_rate, duration=duration)
noise = np.random.normal(0, 0.1, sample_rate * duration)**2
return peak_normalize(signal + noise, -3)
import librosa as lib
import librosa.display
import numpy as np
import matplotlib.pyplot as plt
y = lib.tone(440, sr=22050, length=22050)
plt.plot(y[2000:2100])
plt.show()
spect = lib.stft(y, n_fft=2048, hop_length=512)
spect_db = lib.amplitude_to_db(np.abs(spect))
librosa.display.specshow(spect_db, x_axis='time', y_axis='log')
plt.title('spectrogram')
plt.colorbar()
plt.show()
"""
Created on Sat Jun 5 11:35:07 2021
@author: Admin
"""
sample_duration = librosa.get_duration(y=sample_array, sr=sample_rate)
print("Sample duration: ", sample_duration)
print("Sample: ", sample_array)
print("Sample rate: ", sample_rate)
# Plot the sound file
plt.figure()
plt.subplot(3, 1, 1)
librosa.display.waveplot(sample_array, sr=sample_rate)
plt.title(requested_sample + " - mono")
plt.show()
# Save new file
new_sample = input("Enter new file name: ")
librosa.output.write_wav(new_sample, sample_array, sample_rate)
# Plot cosine signal
plt.figure()
plt.subplot(3, 1, 1)
cosine_map_plot = librosa.tone(1000, sr=sample_rate, duration=0.002)
cosine_map = librosa.tone(1000, sr=sample_rate, duration=sample_duration)
print("Cosine map: ", cosine_map)
# print("Cosine sample rate: ", sample_rate_cosine)
librosa.display.waveplot(cosine_map_plot, sr=sample_rate)
plt.title(requested_sample + " - cosine wave (1KHz)")
plt.show()
# Save Cosine file
cosine_file = input("Enter file name for cosine sample: ")
librosa.output.write_wav(cosine_file, cosine_map, sample_rate)
def y_cqt_110(sr_cqt):
return librosa.tone(110.0, sr_cqt, duration=0.75)
