The difficulty with the problem presented is that "spectral content" is not just a number you can compare to find degrees of proximity. You need to have complex analysis algorithms, comparing multiple parameters depending on the intended application, and define proximity criteria, normally based on statistical methods. This is true even for relatively "simple" sounds (musical instrument samples, or short natural sounds, for example), if the sound is too heterogeneous along time, the task is almost impossible and/or meaningless.

So in theory there could be (and perhaps there is, I don't know) a ready made tool to do, within certain constraints and field of application, the type of cataloging and searching described in the question. But what can be found more easily (normally within the realm of academic and scientific research) are resources and tools to allow to build such a tool to suit one's specific needs after (normally a lot) of experimentation and tuning.

One such tool is Essentia, a very powerful open-source C++ library for audio analysis and audio-based music information retrieval. This library has functions to evaluate dozens of different properties of sound spectra. A look at the Algorithms overview page, particularly the MIR (Musical Information Retrieval) functions is worth a look, to get an idea of the different audio properties that must potentially be considered.

This question at the Music.SE can also help to understand the approach to this kind of problem.

The difficulty with the problem presented is that "spectral content" is not just a number you can compare to find degrees of proximity. You need to have complex analysis algorithms, comparing multiple parameters depending on the intended application, and define proximity criteria, normally based on statistical methods. This is true even for relatively "simple" sounds (musical instrument samples, or short natural sounds, for example), if the sound is too heterogeneous along time, the task is almost impossible and/or meaningless.

So in theory there could be (and perhaps there is, I don't know) a ready made tool to do, within certain constraints and field of application, the type of cataloging and searching described in the question. But what can be found more easily (normally within the realm of academic and scientific research) are resources and tools to allow to build such a tool to suit one's specific needs after (normally a lot) of experimentation and tuning.

One such tool is Essentia, a very powerful open-source C++ library for audio analysis and audio-based music information retrieval. This library has functions to evaluate dozens of different properties of sound spectra. A look at the Algorithms overview page, particularly the MIR (Musical Information Retrieval) functions is worth a look, to get an idea of the different audio properties that must potentially be considered.

This question at the Music.SE can also help to understand the approach to this kind of problem.