1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162 |
- from numpy import mod
- import pandas as pd
- from ChildProject.projects import ChildProject
- from ChildProject.annotations import AnnotationManager
- from ChildProject.metrics import segments_to_grid, conf_matrix, segments_to_annotation
- from pathlib import Path
- def compare_vandam(set1: str, set2: str) :
- speakers = ['CHI', 'OCH', 'FEM', 'MAL']
- project = ChildProject('vandam-data')
- am = AnnotationManager(project)
- #am.read()
- #get segments that intercept between two annotations
- intersection = AnnotationManager.intersection(am.annotations, [set1, set2])
-
- #output directory
- dirName = "outputs/compare/" + set1.replace("/","") + "-" + set2.replace("/","")
- try:
- # Create target Directory
- Path(dirName).mkdir(parents= True)
- print("Directory " , dirName , " Created ")
- except FileExistsError:
- print("Directory " , dirName , " already exists")
-
- #opens output file
- file= open("{0}/{1}-{2}.txt".format(dirName, set1.replace("/",""), set2.replace("/","")),"a")
-
- for speaker in speakers:
- #retrieve contents
- segments = am.get_collapsed_segments(intersection)
- segments = segments[segments['speaker_type'].isin(pd.Series(speaker))]
- ref = segments_to_annotation(segments[segments['set'] == set1], 'speaker_type')
- hyp = segments_to_annotation(segments[segments['set'] == set2], 'speaker_type')
- if __name__ == '__main__':
- #compute metrics
- from pyannote.metrics.detection import DetectionPrecisionRecallFMeasure
- metric = DetectionPrecisionRecallFMeasure()
- detail = metric.compute_components(ref, hyp)
- precision, recall, f = metric.compute_metrics(detail)
-
- #saves metrics to output file
- metric_output = "precision: {0} / recall : {1} / f: {2}\n".format(precision, recall, f)
- file.write(speaker + ": " + metric_output)
- print("Metrics [precision & recall & f] saved! for {0}".format(speaker))
- file.close
- #generates segments
- set1_segm = segments_to_grid(segments[segments['set'] == set1], 0, segments['segment_offset'].max(), 100, 'speaker_type', speakers)
- set2_segm = segments_to_grid(segments[segments['set'] == set2], 0, segments['segment_offset'].max(), 100, 'speaker_type', speakers)
- matrix_df = pd.DataFrame(conf_matrix(set1_segm, set2_segm))
- matrix_df.to_csv("{0}/{1}-{2}-confusion-matrix.csv".format(dirName, set1.replace("/",""), set2.replace("/","")), mode = "x", index=False)
- print("Confusion matrix saved for {0} and {1}!".format(set1, set2))
- compare_vandam('eaf', 'cha')
- compare_vandam('eaf', 'cha/aligned')
|