hi there
I am new in this forum and I need help please.
I have project about recognize english alphabet letters sounds by neural network in matlab
but I don't know what sound features I have to use as a input to a neural network
If I use a samples from the sounds,will it be sufficient to recognize between sounds?
or I must use some other feature
The most difficult issue here is that neural networks are structured to process static data. I.e. look for these types of input values and activate this output. Speech recognition involves both the frequency domain ( magnitudes of spectral components) and the time domain ( the duration of the word ). To use a neural network for this application would require preprocessing of the audio to convert to the frequency domain ( maybe by fft) and then sampling these values for the duration of the word. After this is done, the time and frequency values would need to scaled to account for the speaking rate and sex of the speaker. The resulting matrix of data values ( each value giving the scaled magnitude of a frequency component at a certain scaled time value) could then be presented to the neural net for training.
Anyway that's my best guess