bp/PySoundConcat
1
0
Fork 0
Code Issues 2 Pull Requests Actions Packages Projects Releases Wiki Activity

Creating examples

Browse Source
This commit is contained in:
Sam Perry
2016-04-13 11:30:19 +01:00
parent 5a12ee1fbb
commit cc67314751
4 changed files with 51 additions and 44 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/sppysound/config.py
+16 -16
View File
@@ -1,31 +1,31 @@
# Specify analysis parameters for root mean square analysis.
rms = {
"window_size": 70,
"overlap": 8,
"window_size": 100,
"overlap": 4,
}
f0 = {
"window_size": 2048,
"overlap": 8,
"ratio_threshold": 0.1
"ratio_threshold": 0.4
}
# Specify analysis parameters for variance analysis.
variance = {
"window_size": 70,
"overlap": 8
"window_size": 100,
"overlap": 4
}
# Specify analysis parameters for temporal kurtosis analysis.
kurtosis = {
"window_size": 70,
"overlap": 8
"window_size": 100,
"overlap": 4
}
# Specify analysis parameters for temporal skewness analysis.
skewness = {
"window_size": 70,
"overlap": 8
"window_size": 100,
"overlap": 4
}
# Specify analysis parameters for FFT analysis.
@@ -48,7 +48,7 @@ matcher_weightings = {
"spcflux" : 2.,
"spccf" : 2.,
"spcflatness": 3.,
"zerox" : 0.,
"zerox" : 1.,
"rms" : 1,
"peak": 2.,
"centroid": 1.,
@@ -88,11 +88,11 @@ analysis = {
matcher = {
# Force the re-matching of analyses
"rematch": False,
"grain_size": 70,
"overlap": 8,
"grain_size": 100,
"overlap": 4,
# Defines the number of matches to keep for synthesis. Note that this must
# also be specified in the synthesis config
"match_quantity": 20,
"match_quantity": 10,
# Choose the algorithm used to perform matching. kdtree is recommended for
# larger datasets.
"method": 'kdtree'
@@ -109,14 +109,14 @@ synthesizer = {
"enforce_f0": True,
# Specify the ratio limit that is the grain can be modified by.
"enf_f0_ratio_limit": 100.,
"grain_size": 70,
"overlap": 8,
"grain_size": 100,
"overlap": 4,
# Normalize output, avoid clipping of final output by scaling the final
# frames.
"normalize" : True,
# Defines the number of potential grains to choose from matches when
# synthesizing output.
"match_quantity": 20
"match_quantity": 10
}
output_file = {
src/sppysound/database.py
+13 -12
View File
@@ -426,15 +426,15 @@ class Matcher:
analysis_formatting = self.analysis_dict[analysis]
target_data, s = target_entry.analysis_data_grains(target_times, analysis, format=analysis_formatting)
target_data = target_data ** weightings[analysis]
target_data **= weightings[analysis]
all_target_analyses[i] = target_data
# nan_columns = np.all(np.isnan(all_target_analyses), axis=0)
# all_target_analyses[:, nan_columns] = 0.
nan_columns = np.all(np.isnan(all_target_analyses), axis=0)
all_target_analyses[:, nan_columns] = 0.
# Impute values for Nans
# all_target_analyses = imp.fit_transform(all_target_analyses)
all_target_analyses[np.isnan(all_target_analyses)] = np.inf
all_target_analyses = imp.fit_transform(all_target_analyses)
# all_target_analyses[np.isnan(all_target_analyses)] = np.inf
for sind, source_entry in enumerate(self.source_db.analysed_audio):
self.logger.info("K-d Tree Matching: {0} to {1}".format(source_entry.name, target_entry.name))
@@ -450,16 +450,16 @@ class Matcher:
analysis_formatting = self.analysis_dict[analysis]
source_data, s = source_entry.analysis_data_grains(source_times, analysis, format=analysis_formatting)
source_data = source_data ** weightings[analysis]
source_data **= weightings[analysis]
all_source_analyses[i] = source_data
# Impute values for Nans
# nan_columns = np.all(np.isnan(all_source_analyses), axis=0)
# all_source_analyses[:, nan_columns] = 0.
# all_source_analyses = imp.fit_transform(all_source_analyses)
nan_columns = np.all(np.isnan(all_source_analyses), axis=0)
all_source_analyses[:, nan_columns] = 0.
all_source_analyses = imp.fit_transform(all_source_analyses)
all_source_analyses[np.isnan(all_source_analyses)] = np.inf
# all_source_analyses[np.isnan(all_source_analyses)] = np.inf
source_tree = spatial.cKDTree(all_source_analyses.T, leafsize=100)
results_vals, results_inds = source_tree.query(all_target_analyses.T, k=self.match_quantity, p=2)
@@ -736,8 +736,9 @@ class Matcher:
)
if not np.all(np.any(x, axis=1)):
pdb.set_trace()
raise ValueError("Not all match indexes have a corresponding sample index. This shouldn't happen...")
raise ValueError("Not all match indexes have a corresponding sample index. This shouldn't happen...\n"
"Check that all database path arguments are correct then try re-running with the --rematch and --reanalyse flags.\n"
"If this does'nt work, delete the audio and data directories in all databases and try again...")
x = x.reshape(mi_shape[0], mi_shape[1], x.shape[1])
x = np.argmax(x, axis=2)
src/sppysound/docs/descriptor_defs.rst
+16 -16
View File
@@ -49,7 +49,7 @@ the fundamental period.
In order to improve the accuracy of peak detection, parabolic interpolation is
used to estimate the peak's location with greater accuracy by using the peak
correlation and it's two closes neighbour's values to estimate the fractional
correlation and it's two closest neighbour's values to estimate the fractional
peak value.
The method for parabolic interpolation is defined as:
@@ -78,7 +78,7 @@ Fourier Transform for windows of a signal. The full description of this
transform is outside the scope of this project, however it should be understood
that this analysis provides a description of the spectral content of a windowed
signal. By applying the transform, a number of bins of size :math:`K` are
calculated that detail the sine and cosine apmlitudes required to reconstruct
calculated that detail the sine and cosine amplitudes required to reconstruct
the signal. The calculation of the STFT is defined as:
.. math::
@@ -90,9 +90,9 @@ Ref: :cite:`lerch2012itaca`
Harmonic Ratio
~~~~~~~~~~~~~~
The harmonic ratio can be used to differentiate between noisy and periodic
signals. higher values suggest that the signal is more periodic (such as a sine
signals. Higher values suggest that the signal is more periodic (such as a sine
wave) and lower values represent less periodicity. This can be used as a form
of confidence measure in determining the validity of F0 values. it is
of confidence measure in determining the validity of F0 values. It is
calculated as part of the F0 estimation algorithm as:
.. math::
@@ -113,7 +113,7 @@ Ref: :cite:`lerch2012itaca`
Peak Amplitude
~~~~~~~~~~~~~~
Peak amplitude measures the highest peak in the absoulte signal. it is
Peak amplitude measures the highest peak in the absolute signal. It is
calculated as:
.. math::
@@ -136,10 +136,10 @@ Ref: :cite:`lerch2012itaca`
Spectral Centroid
~~~~~~~~~~~~~~~~~
The spectral centroid measure the center of gravity accross frequency bins to
determine the central point accross the spectral content of the frame. High
value sindicate that the spectral content is centered in higher frequencies and
lower value indicate a lower center. The spectral centroid is calculated as:
The spectral centroid measure the centre of gravity across frequency bins to
determine the central point across the spectral content of the frame. High
values indicate that the spectral content is centred in higher frequencies and
lower value indicate a lower centre. The spectral centroid is calculated as:
.. math::
SC(n) = \frac{\sum_{k=0}^{K/2-1} k \cdot | X(k,n) | ^2}{\sum_{k=0}^{K/2-1} | X(k,n) | ^2}
@@ -151,11 +151,11 @@ Ref: :cite:`lerch2012itaca`
Spectral Crest Factor
~~~~~~~~~~~~~~~~~~~~~
The spectral crest factor can be used as a mesure of tonalness of the signal.
it is calculated by taking the maximum magnitude and dividing by the sum of
The spectral crest factor can be used as a measure of tonalness of the signal.
It is calculated by taking the maximum magnitude and dividing by the sum of
magnitudes.
This differntiates between flat spectrums and sinusoidal spectrums. (low values
represnting the former and high values representing the latter.)
This differentiates between flat spectrums and sinusoidal spectrums. (low values
representing the former and high values representing the latter.)
.. math::
SCF = \frac{ \max_{0 \leq k \leq K/2-1} \{| X(k,n) | \}}{\sum_{k=0}^{K/2-1} | X(k,n) | }
@@ -180,7 +180,7 @@ Spectral Flux
Spectral flux is a measure of change between consecutive frames. It calculates
the average difference between frames to differentiate between adjacent frames
that are largely dissimilar (suggesting a non-stationary section of signal) and
similiar frames (that suggests a steady state signal). It is calculated as:
similar frames (that suggests a steady state signal). It is calculated as:
.. math::
SF(n) = \frac{\sqrt{\sum_{k=0}^{K/2-1} \Big( | X(k,n) | - | X(k,n-1) | \Big)^2
@@ -213,8 +213,8 @@ Ref: :cite:`lerch2012itaca`
Zero-Crossing
~~~~~~~~~~~~~
The zero-crossing rate counts the number of times a signal's value changes from
positive to negative in a frame. it is relevant to determining the noisiness of
a signal, as noisy signals will pass from positive to negative more frequenctly
positive to negative in a frame. It is relevant to determining the noisiness of
a signal, as noisy signals will pass from positive to negative more frequently
than period signals. It is calculated as:
.. math::
src/tests/config.py
+6
View File
@@ -3,6 +3,12 @@ rms = {
"overlap": 16,
}
f0 = {
"window_size": 2048,
"overlap": 8,
"ratio_threshold": 0.0
}
variance = {
"window_size": 130,
"overlap": 16