def process_csv_data(i, file_path, cqt_length):
print('[{}] Processing CSV data...'.format(i))
result = np.array([])
with open(file_path) as csvfile:
reader = csv.reader(csvfile, delimiter='\t')
# Skip header
next(reader)
row_count = sum(1 for row in reader)
sets = np.zeros([row_count, 3])
csvfile.seek(0)
next(reader)
for i, row in enumerate(reader):
if len(row) != 3:
continue
onset = row[0]
offset = row[1]
midi_pitch = row[2]
sets[i] = [onset, offset, midi_pitch]
result = np.zeros([PITCH_RANGE * 2, cqt_length])
time_stamps = frames_to_time(range(0, cqt_length + 1),
sr=TARGET_SAMPLE_RATE,
hop_length=HOP_LENGTH)
process_csv_data_jit(sets, time_stamps, cqt_length, result)
return result
def index():
form = Params(request.form)
# if user submit the form
if (request.method == 'POST' and form.validate()):
# Get data from form
extratorAlgorithm, filename, centroids_number, musicPath = getFormData(
)
# Extract features, then normalize
matrix = getFeatureMatrix(extratorAlgorithm, filename)
matrix_norm = normalize(matrix)
# Get the centroids and calculate the distance to all frames
centroids, distancesCentroid = getCentroids(matrix_norm,
centroids_number)
distancesCentroid = np.array(distancesCentroid)
# Define the number of closest frames from centroids to get
centroidClosestPoints = 5
# Get the ID of frames closest to each centroid
closest_frames_idx = distancesCentroid.argsort(axis=0).T.tolist()
centroid_frame_counts = np.bincount(
np.argmin(distancesCentroid, axis=1))
# Get the time from each frame
tempos = frames_to_time(closest_frames_idx,
sr=44100,
hop_length=1024,
n_fft=2048)
tempos = tempos.tolist()
# Apply some transformations
pca = reduceDimensionality(matrix_norm)
matrix_norm = pca.transform(matrix_norm)
centroids = pca.transform(centroids)
# Generate and encode graph to send to view
graph = Graph(
matrix_norm,
centroids,
filename,
centroid_frame_count=centroid_frame_counts).generateGraph()
print tempos
# Send all structures to the view
return render_template('index.html',
graph=graph,
form=form,
musicPath=musicPath,
tempos=tempos)
# Generate an empty graphic
graph = Graph().generateGraph()
return render_template('index.html', graph=graph, form=form)
def __coord_time(n, sr=22050, hop_length=512, **_kwargs):
"""Get time coordinates from frames"""
return core.frames_to_time(np.arange(n + 1), sr=sr, hop_length=hop_length)
def onset_detect(
y=None,
sr=22050,
onset_envelope=None,
hop_length=512,
backtrack=False,
energy=None,
units="frames",
normalize=True,
**kwargs
):
"""Basic onset detector. Locate note onset events by picking peaks in an
onset strength envelope. Modified from `librosa.onset.onset_detect` to add a
`normalize` flag.
The `peak_pick` parameters were chosen by large-scale hyper-parameter
optimization over the dataset provided by [1]_.
.. [1] https://github.com/CPJKU/onset_db
Parameters
----------
y : np.ndarray [shape=(n,)]
audio time series
sr : number > 0 [scalar]
sampling rate of `y`
onset_envelope : np.ndarray [shape=(m,)]
(optional) pre-computed onset strength envelope
hop_length : int > 0 [scalar]
hop length (in samples)
units : {'frames', 'samples', 'time'}
The units to encode detected onset events in.
By default, 'frames' are used.
backtrack : bool
If `True`, detected onset events are backtracked to the nearest
preceding minimum of `energy`.
This is primarily useful when using onsets as slice points for segmentation.
energy : np.ndarray [shape=(m,)] (optional)
An energy function to use for backtracking detected onset events.
If none is provided, then `onset_envelope` is used.
noramlize : bool (optional)
If `True`, normalize the onset envelope before peak picking. By default
this parameter is `True`.
kwargs : additional keyword arguments
Additional parameters for peak picking.
See `librosa.util.peak_pick` for details.
Returns
-------
onsets : np.ndarray [shape=(n_onsets,)]
estimated positions of detected onsets, in whichever units
are specified. By default, frame indices.
.. note::
If no onset strength could be detected, onset_detect returns
an empty list.
Raises
------
ParameterError
if neither `y` nor `onsets` are provided
or if `units` is not one of 'frames', 'samples', or 'time'
"""
# First, get the frame->beat strength profile if we don't already have one
if onset_envelope is None:
if y is None:
raise ParameterError("y or onset_envelope must be provided")
onset_envelope = onset_strength(y=y, sr=sr, hop_length=hop_length)
# Shift onset envelope up to be non-negative
# (a common normalization step to make the threshold more consistent)
onset_envelope -= onset_envelope.min()
# Do we have any onsets to grab?
if not onset_envelope.any():
return np.array([], dtype=np.int)
if normalize:
# Normalize onset strength function to [0, 1] range
onset_envelope /= onset_envelope.max()
# These parameter settings found by large-scale search
kwargs.setdefault("pre_max", 0.03 * sr // hop_length) # 30ms
kwargs.setdefault("post_max", 0.00 * sr // hop_length + 1) # 0ms
kwargs.setdefault("pre_avg", 0.10 * sr // hop_length) # 100ms
kwargs.setdefault("post_avg", 0.10 * sr // hop_length + 1) # 100ms
kwargs.setdefault("wait", 0.03 * sr // hop_length) # 30ms
kwargs.setdefault("delta", 0.07)
# Peak pick the onset envelope
onsets = util.peak_pick(onset_envelope, **kwargs)
# Optionally backtrack the events
if backtrack:
if energy is None:
energy = onset_envelope
onsets = onset_backtrack(onsets, energy)
if units == "frames":
pass
elif units == "samples":
onsets = core.frames_to_samples(onsets, hop_length=hop_length)
elif units == "time":
onsets = core.frames_to_time(onsets, hop_length=hop_length, sr=sr)
else:
raise ParameterError("Invalid unit type: {}".format(units))
return onsets
3464.84375, 3468.75, 3472.65625, 3476.5625, 3480.46875, 3484.375,
3488.28125, 3492.1875, 3496.09375, 3500.0, 3503.90625, 3507.8125,
3511.71875, 3515.625, 3519.53125, 3523.4375, 3527.34375, 3531.25,
3535.15625, 3539.0625, 3542.96875, 3546.875, 3550.78125, 3554.6875,
3558.59375, 3562.5, 3566.40625, 3570.3125, 3574.21875, 3578.125,
3582.03125, 3585.9375, 3589.84375, 3593.75, 3597.65625, 3601.5625,
3605.46875, 3609.375, 3613.28125, 3617.1875, 3621.09375, 3625.0,
3628.90625, 3632.8125, 3636.71875, 3640.625, 3644.53125, 3648.4375,
3652.34375, 3656.25, 3660.15625, 3664.0625, 3667.96875, 3671.875,
3675.78125, 3679.6875, 3683.59375, 3687.5, 3691.40625, 3695.3125,
3699.21875, 3703.125, 3707.03125, 3710.9375, 3714.84375, 3718.75,
3722.65625, 3726.5625, 3730.46875, 3734.375, 3738.28125, 3742.1875,
3746.09375
]
frequency_strength_thr = 1e2
times_of_bins = lambda hm_steps: frames_to_time(range(
0, hm_steps), sample_rate, fft_hop_len, fft_bins)
zscore_scale = True
minmax_scale = False
log_scale = False
data_path = 'data'
dev_ratio = 0
## model params
timestep_size = len(frequencies_to_pick)
in_size = timestep_size
out_size = timestep_size
def frame_level_predict(model_name, filename, cache=True, plot=False):
"""
Predict Voice Activity Regions at a Frame Level for a given song.
For each frame of the MFCC a Voice Detection Probability is predicted, then the output have shape: (n_frames, 1)
:param model_name: name of the trained model
:param filename: path to the music file to be predicted
:param cache: flag to optimize heavy operations with caching in disk
:param plot: flag to plot MFCCs and SVD in an aligned plot if GUI available.
:return: (Time, Predictions): SVD probabilities at frame level with time markings
"""
audio_name = filename.parts[-1]
audio_name_prefix = '.'.join(filename.parts[:-1])
serialized_filename = PREDICTIONS_DIR / '{}.{}.{}.csv'.format(
audio_name_prefix, audio_name, model_name)
mel = process_single_audio(filename, cache=cache)
try:
if not cache:
raise IOError
data = np.loadtxt(serialized_filename, delimiter=',')
time = data[0]
frame_level_y_pred = data[1]
print("info: loaded serialized prediction")
except Exception:
# transform raw predictions to frame level
y_pred = predict_song(model_name, filename, cache=cache)
aligned_y_pred = [[] for _ in range(mel.shape[1])]
for first_frame_idx, window_prediction in enumerate(y_pred):
# for each prediction
for offset, frame_prediction in enumerate(window_prediction):
# accumulate overlapped predictions in a list
aligned_y_pred[first_frame_idx + offset].append(
frame_prediction[0])
frame_level_y_pred = []
for _, predictions in enumerate(aligned_y_pred[:-1]):
# reduce the overlapped predictions to a single value
frame_level_y_pred.append(min(predictions))
time = frames_to_time(range(len(frame_level_y_pred)),
sr=SR_HPSS,
n_fft=N_FFT_HPSS_2,
hop_length=N_HOP_HPSS_2)
np.savetxt(serialized_filename,
np.asarray((time, frame_level_y_pred)),
delimiter=",")
print("info: saved serialized prediction")
if plot:
import matplotlib.pyplot as plt
import librosa.display
# plot stacked MFCCs
plt.figure(figsize=(14, 5))
plt.subplot(211)
librosa.display.specshow(mel,
sr=SR_HPSS,
x_axis='time',
y_axis='hz',
hop_length=N_HOP_HPSS_2)
# plot frame level predictions
plt.subplot(313)
plt.plot(time, frame_level_y_pred)
plt.xlabel("Time")
plt.ylabel("Singing Voice Activation")
plt.show()
print("info: plotted")
print('info: done')
return time, frame_level_y_pred