5a1dba682a5eccd4f0854553c07cd2b21909f1ec,pyannote/audio/labeling/tasks/resegmentation.py,Resegmentation,_decode,#Resegmentation#Any#Any#Any#Any#,312
Before Change
// K = 1 <~~> only non-speech
// K = 2 <~~> just one speaker
if K < 3:
return hypothesis
// sequence of most likely speaker index
// (even when non-speech is the most likely class)
best_speaker_indices = np.argmax(scores.data[:, 1:], axis=1) + 1
// reconstruct annotation
new_hypothesis = one_hot_decoding(
After Change
N, K = scores.data.shape
if self.allow_overlap:
active_speakers = np.exp(scores.data > 0.5)
else:
if self.lock_speech:
active_speakers = np.argmax(scores.data, axis=1)
In pattern: SUPERPATTERN
Frequency: 3
Non-data size: 5
Instances
Project Name: pyannote/pyannote-audio
Commit Name: 5a1dba682a5eccd4f0854553c07cd2b21909f1ec
Time: 2020-03-17
Author: bredin@limsi.fr
File Name: pyannote/audio/labeling/tasks/resegmentation.py
Class Name: Resegmentation
Method Name: _decode
Project Name: PIQuIL/QuCumber
Commit Name: aacf14504b82087378db638d8f0463bc9933cabd
Time: 2018-05-22
Author: emerali@users.noreply.github.com
File Name: benchmarks/python/rbm.py
Class Name: RBM
Method Name: ln1pexp
Project Name: scipy/scipy
Commit Name: b0f34cdfd3ec8e41521cfd1b4a19c8dabb51fbbf
Time: 2020-02-15
Author: pvanmulbregt@users.noreply.github.com
File Name: scipy/stats/_ksstats.py
Class Name:
Method Name: _kolmogn