For instance, instead of using "8 sawtooth oscillators and each one is assigned 50 notes from the 400 digits", the composer could have use 2 samples of babies crying and 3 of babies laughing, with some more digits assigned to amplitude and duration of each sample.
The result would sound completely different, and yet hold just as valid a claim to being "The Sounds of Pi".
Or maybe you could assign certain combinations of digits to samples of instruments from an orchestra, or a didgeridoo, or samples of people reading from Shakespeare or a comic, to sounds of car crashes, or directly to frequency and amplitude variation, or any combination of the above. These assignments could be "random", or chosen by Pi itself, or through some algorithm, or by hand by the composer.
The above doesn't even touch on filtering or effects that could be applied... which themselves could be controlled and/or assigned by digits of Pi.
There are an infinite number of ways you could represent any given sequence of numbers as sound, which are really only limited by the composer's imagination.
So what does Pi really sound like? It doesn't sound like anything, or it sounds like anything, depending on what choices one makes.