audio recognition not speech

It’s possible, but very difficult. You’d have to tweak the underlying Sphinx recognizer.

Sphinx is normally set up as a “cepstral” recognizer. That’s an approach where it tries to recognize the process that makes a sound, when the process (for example, the passage of sound through the throat, mouth, teeth, and lips) is more important than the sound itself.

You’re after the signature of the sound itself. I remember a paper from Avery Wang on how the “Shazam” music recognizer worked that described this.

To do it with Sphinx, you need to recognize the basic sound, so you need to set up the Sphinx front end to use “energy” features, not “cepstral” and delta cepstral features. Then you need to collect some sounds of a ball hitting a wall and train a Sphinx acoustic model to recognize that.

Good luck, and have fun.