Implement an audio filter, using some filter algorithm that does not currently exist as an MSP or Pd object.
Implement some type of useful "spatialization" of an audio signal based on x,y coordinate information for location of the sound (received as floats or signals in other inlets), and sending 2 or 4 signals out to go to different speakers. Ways of achieving this illusion of a "virtual location" might include delays, intensity differences, "Ambisonic" techniques, etc.
Granularize a prerecorded signal that is being stored in RAM in a separate buffer~ object.
Provide useful numerical data about some aspect of an audio signal (e.g., its rms amplitude, its peak amplitude, its number of zero crossings, its strongest frequency region, etc.).
Compress, compand, or limit a signal in some other way that has not yet been implemented as an MSP or Pd object.
Make an envelope follower that tracks some feature of an incoming signal (such as its peak amplitude) and sends out a new signal that interpolates between the tracked values.
Implement a synthesis method described in the audio engineering or computer music literature (e.g., Karplus-Strong, linear predictive coding, scanned synthesis, etc.).
Devise a new means of audio processing that has not yet been implemented (e.g., periodically mixing a signal with a reversed version of itself).
For additional ideas, follow some of the links listed on the Useful Links page.