I am trying to calculate the mfcc of an audio sample of 5sec duration. How can I frame this audio sample. Is there any criterion for selecting the frame size and number of frames? What all change will be there if the audio file is 30sec long?
Usually in speech recognition frame size is around 16 milliseconds (0.016 second), which corresponds to average length of an audio signal during which it is stationary (not to much variation). Regarding the overlap I suggest you to start with 50%, you can then try to play with it, but it should not change too much.
I do not really understand part of you question asking what will change if the file is 30 seconds long. Obviously frame length stays the same and only number of frames will increase.