Hi Abraham,
In order to separate out the individual frequencies present in an audio signal, MFCCs are computed over the signals in the audio. A single vector of MFCC is produced as the result of Short-Time Fourier Transfrom (STFT). MFCCs are computed over a frame of 25ms, with a stride of 10 ms between each frame. Therefore, you will get 100 vectors per second of speech, which gives you a matrix of shape (100, 13) for the resultant MFCC.
To sum it up, the 13 MFCCs are the 13 mel-frequency cepstral coefficients for the corresponding frame of the speech file. For any more help on functions on MFCC, you can refer to the link here.
