The feature data of this homework are the same as homework 1 part 2. There are many variable length utterances comprised of 40 dimensional timesteps. You can use recurrent layers and their variants. You may also use convolutional layers if you so choose.
Your task will be to take the sequence of features and map them to a sequence of phonemes. Keep in mind the number of phonemes will not equal the number of timesteps. Refer to the write-up for more information.