"""
dataCompilation
===============
.. autosummary::
:toctree: generated/
data_compilation
"""
import numpy as np
import librosa
from pyampact.performance import estimate_perceptual_parameters
from pyampact.alignmentUtils import f0_est_weighted_sum_spec
from pyampact.symbolic import Score
__all__ = [
"data_compilation"
]
[docs]
def data_compilation(y, original_sr, nmat, piece, audio_file_path, output_path='output.mei'):
"""
This function takes the results of the alignment and the note matrix and compiles the data into a JSON object
that can be used to insert the audio analysis into the score.
Parameters
----------
nmat : np.ndarray
The note matrix containing information about notes, including their timing and duration.
audio_file : str
The path to the audio file associated with the performance data.
piece : Score
An instantiation of the original Score object containing the data input for the musical piece.
output_path : str, optional
The file path for the output MEI file. Defaults to './output.mei'.
Returns
-------
nmat : The note matrix with performance data appended.
json_data : A JSON object containing the compiled data.
xml_data : XML data representing the MEI output.
"""
# y, original_sr = librosa.load(audio_file)
# print(original_sr)
all_note_vals = []
all_note_ids = []
for key, df in nmat.items():
df = df.drop(columns=['MEASURE', 'ONSET', 'DURATION', 'PART', 'MIDI'])
midiList = np.array(nmat[key]['MIDI'])
loc = 1
f0 = []
pwr = []
t = []
M = []
xf = []
note_vals = []
note_ids = []
for loc in range(len(df)):
# Estimate f0 for a matrix (or vector) of amplitudes and frequencies
if df['OFFSET_SEC'].iloc[loc] - df['ONSET_SEC'].iloc[loc] > 0:
[f0, pwr, t, M, xf] = f0_est_weighted_sum_spec(
df['ONSET_SEC'].iloc[loc], df['OFFSET_SEC'].iloc[loc], midiList[loc], y, original_sr)
# Estimate note-wise perceptual values
# if flag = M1, call pass in original, if M2 pass in reconstructed
note_vals.append(estimate_perceptual_parameters(
f0, pwr, M, original_sr, 256, 1))
note_ids.append(nmat[key].index[loc])
else:
print([df['ONSET_SEC'].iloc[loc], df['OFFSET_SEC'].iloc[loc]])
all_note_vals.append(note_vals)
all_note_ids.append(note_ids)
loc = 0
for key, df in nmat.items():
df['f0Vals'] = [all_note_vals[loc][i]['f0_vals'] for i in range(len(df))]
df['meanf0'] = [np.mean(vals) for vals in df['f0Vals']]
df['ppitch1'] = [all_note_vals[loc][i]['ppitch'][0] for i in range(len(df))]
df['ppitch2'] = [all_note_vals[loc][i]['ppitch'][1] for i in range(len(df))]
df['jitter'] = [all_note_vals[loc][i]['jitter'] for i in range(len(df))]
df['vibratoDepth'] = [all_note_vals[loc][i]['vibrato_depth'] for i in range(len(df))]
df['vibratoRate'] = [all_note_vals[loc][i]['vibrato_rate'] for i in range(len(df))]
df['pwrVals'] = [all_note_vals[loc][i]['pwr_vals'] for i in range(len(df))]
df['meanPwr'] = [np.mean(vals) for vals in df['pwrVals']]
df['shimmer'] = [all_note_vals[loc][i]['shimmer'] for i in range(len(df))]
df['specCentVals'] = [all_note_vals[loc][i]['spec_centroid'] for i in range(len(df))]
df['meanSpecCent'] = [np.mean(vals) for vals in df['specCentVals']]
df['specBandwidthVals'] = [all_note_vals[loc][i]['spec_bandwidth'] for i in range(len(df))]
df['meanSpecBandwidth'] = [np.mean(vals) for vals in df['specBandwidthVals']]
df['specContrastVals'] = [all_note_vals[loc][i]['spec_contrast'] for i in range(len(df))]
df['meanSpecContrast'] = [np.mean(vals) for vals in df['specContrastVals']]
df['specFlatnessVals'] = [all_note_vals[loc][i]['spec_flatness'] for i in range(len(df))]
df['meanSpecFlatness'] = [np.mean(vals) for vals in df['specFlatnessVals']]
df['specRolloffVals'] = [all_note_vals[loc][i]['spec_rolloff'] for i in range(len(df))]
df['meanSpecRolloff'] = [np.mean(vals) for vals in df['specRolloffVals']]
loc += 1
def convert_nmat_for_export(nmat):
"""
Converts columns with list values into stringified versions for export (e.g., MEI, CSV, JSON).
Parameters
----------
nmat : dict of DataFrames
The processed note matrix with internal Python lists.
Returns
-------
export_nmat : dict of DataFrames
A new dictionary where list-valued columns are stringified.
"""
list_columns = [
'f0Vals', 'pwrVals', 'specCentVals', 'specBandwidthVals',
'specContrastVals', 'specFlatnessVals', 'specRolloffVals'
]
export_nmat = {}
for part, df in nmat.items():
df_copy = df.copy()
for col in list_columns:
if col in df_copy.columns:
df_copy[col] = df_copy[col].apply(lambda x: str(x))
export_nmat[part] = df_copy
return export_nmat
nmat_export = convert_nmat_for_export(nmat)
if getattr(piece, 'fileExtension') != 'csv':
output_path = './output.mei'
fileOutput = piece.insertAudioAnalysis(
output_path=output_path, data=nmat_export, mimetype='audio/aiff', target=audio_file_path)
else:
# Save each part of nmat to a CSV file
for part, df in nmat_export.items():
# Construct the filename using the filepath, label, and part
filename = f"./output.csv"
# Save the DataFrame to CSV
fileOutput = df.to_csv(filename)
print(f"Saved {filename}")
return nmat, fileOutput