Agree with Tim - remember that spectrograms usually present a relativley long time window. So if you want a single value for amplitude/frequency at a moment in time then a very small window in probably the best to use in your analysis.
Taking into account that the audio signal is a non-stationary random process, you have two possibilities: comparing short-term power spectra of signals (10-20 ms) or comparing the average power in one-third octave bands at large time intervals. In both cases, the amplitude estimation at a certain frequency is not possible.