Update task.md

2/task.md

@@ -8,7 +8,7 @@ To save time you will be provided with open implementation which you're free to
 The preparation of train / test datasets are full responsibility of a student.
 
 Task:
-*  Learn how to extract MFCC + FBANK features using librosa and and new feature extractor in laughter_prediction.feature_extractors.py
+*  Learn how to extract MFCC + FBANK features using librosa and add new feature extractor in laughter_prediction.feature_extractors.py
 *  Implement a new predictor in lauhter_prediction.predictors.py based on RNN architecture. Use separate loss function for MFCC features only and for the final output. Example can be found here: https://keras.io/getting-started/functional-api-guide/ at Multi-input and multi-output models section
 *  Add noise to dataset using robustiier from the first homework or download dataset with added noise from <coming soon> and compare prediction accuracy and clean data and data with added noise. Which dataset you'd use for train and which for test?
 *  Publish notebook with framewise AUC and plot a couple of audio samples with true laughter intervals with predicted laughter probability