To briefly comment on RBJ's
The Short-Time Fourier Transform (STFT) is an analysis tool that repeatedly applies the Discrete Fourier Transform (DFT or FFT) to short snippets of a longer signal, while sliding the window that selects the snippet of the longer signal that the STFT operates on. The STFT is a function of two independent variables, frequency in the DFT and the time of the center of the sliding window.
I think "repeatedly applies ... operates on" should stand in its own sentence and start like "Its standard implementation", and end with "but it can also be done with strided convolutions". That implementation also lacks a corrective step for some purposes.
"Windowed Fourier transform" is one (and the more common) interpretation, but not the only one - more generally it's a time-frequency decomposition via complex bandpass filters. STFT also isn't exclusively discrete, and even then "frequency in the DFT" is too restrictive, even if not going with DFT freqs is a bad idea.
dsp.meta
or do you just check? $\endgroup$