Any number of fellow-students wanted to paint O'Keeffe, whose strong, clean features and intensely expressive hands and eyes would so fascinate Stieglitz a few years later.
(A) Clean feature vectors.
"That clean feature database is just essential.
For each of the 13 static features, we replace noisy feature sequences with clean feature sequences (RwC: replace with clean).
In the first set of experiments, noisy feature sequences are replaced by the corresponding clean feature sequences.
We train the DNN to predict both the clean features and their time derivatives from distorted features.
The dictionary for clean feature space and the dictionary for reverberant feature space were jointly trained using the stereo data in order to have common representation coefficients.
Another reason is that it may be useful to explicitly recover the clean features rather than let the DNN acoustic model automatically discover useful features for speech recognition.
In [25], a special case of recurrent neural networks, called long short-term memory (LSTM), is used to map reverberant features to clean features.
Li et al. [12] propose to train a joint sparse transformation to estimate the clean feature vector from the reverberant feature vector.
