Update datafiles.md

spokenlanguage · Dec 11, 2019 · ac86f8b · ac86f8b
1 parent 9ea106d
commit ac86f8b
Showing 1 changed file with 8 additions and 40 deletions.
diff --git a/platalea/analysis/datafiles.md b/platalea/analysis/datafiles.md
@@ -6,49 +6,17 @@ For analyzing the VGS model we are using the data files described below. For eac
 ```
 global_input.pkl
 {'audio_id': <class 'numpy.ndarray'>, 
- 'ipa': <class 'numpy.ndarray'>, 
- 'text': <class 'numpy.ndarray'>, 
- 'audio': <class 'numpy.ndarray'>}
+ 'audio': <class 'numpy.ndarray'>, 
+ 'text': <class 'numpy.ndarray'>}
 
-global_trained.pkl
-{'conv': <class 'numpy.ndarray'>, 
- 'rnn0': <class 'numpy.ndarray'>, 
- 'rnn1': <class 'numpy.ndarray'>, 
- 'rnn2': <class 'numpy.ndarray'>, 
- 'rnn3': <class 'numpy.ndarray'>, 
- 'att': <class 'numpy.ndarray'>}
+global_trained_LAYER.pkl
+{'LAYER': <class 'numpy.ndarray'> }
 
-global_random.pkl
-{'conv': <class 'numpy.ndarray'>, 
- 'rnn0': <class 'numpy.ndarray'>, 
- 'rnn1': <class 'numpy.ndarray'>, 
- 'rnn2': <class 'numpy.ndarray'>, 
- 'rnn3': <class 'numpy.ndarray'>, 
- 'att': <class 'numpy.ndarray'>}
-```
-
-## Local
-
-```
-local_input.pkl
-{'features': <class 'numpy.ndarray'>, 'labels': <class 'numpy.ndarray'>}
+global_random_LAYER.pkl
+{'LAYER': <class 'numpy.ndarray'> }
 
-local_trained.pkl
-{'conv': {'features': <class 'numpy.ndarray'>, 'labels': <class 'numpy.ndarray'>}, 
- 'rnn0': {'features': <class 'numpy.ndarray'>, 'labels': <class 'numpy.ndarray'>}, 
- 'rnn1': {'features': <class 'numpy.ndarray'>, 'labels': <class 'numpy.ndarray'>}, 
- 'rnn2': {'features': <class 'numpy.ndarray'>, 'labels': <class 'numpy.ndarray'>}, 
- 'rnn3': {'features': <class 'numpy.ndarray'>, 'labels': <class 'numpy.ndarray'>}}
+downsampling_factors.json
 
-local_random.pkl
-{'conv': {'features': <class 'numpy.ndarray'>, 'labels': <class 'numpy.ndarray'>}, 
- 'rnn0': {'features': <class 'numpy.ndarray'>, 'labels': <class 'numpy.ndarray'>}, 
- 'rnn1': {'features': <class 'numpy.ndarray'>, 'labels': <class 'numpy.ndarray'>}, 
- 'rnn2': {'features': <class 'numpy.ndarray'>, 'labels': <class 'numpy.ndarray'>}, 
- 'rnn3': {'features': <class 'numpy.ndarray'>, 'labels': <class 'numpy.ndarray'>}}
+fa.json
 
 ```
-
-
-In these dictionaries 'features' correspond to inputs or activations for a single frame or timestep. The 'labels' array contains the corresponding phoneme labels, based on forced alignment between input audio and its transcription. Note that in this dataset some utterances have been filtered out due to the force alignment failing.
-