TRIVIAL QUESTION: I am currently working with some audio data of speech utterances. I am attempting to perform classification on the data based on the phonemes. This means that I manually label the audio signal based on the phonemes (the waveform is very distinct for each phoneme) and then I attempt to classify 40ms windows of the data using a Neural Network. I extract some 15 features(will increase features) from each one of the 40ms frames and then train an ANN based on those features. sampling rate/FS = 44.1kHz for all audio samples
Q1 : What is an appropriate normalization technique to employ? (I am normalizing to avoid issues with having multiple speakers)
Q2: Will I need to apply a Hanning window, Hamming window, etc. on the extracted 40ms frames/windows of audio? Why or why not?