After a couple of hours of using strace to try to understand what Nashville was doing, and looking at the "base_audio" directory in the MusicPal kernel source directory, I have finally got it to make a noise.
Basically you have to use /dev/audio and send some ioctl's to initialise the hardware, and set up the graphic equaliser. You can then write raw data to /dev/audio just like you would with OSS. In fact, I used the OSS sinegen.c example program as the basis and just altered the ioctls. http://www.musicpal.webhop.net/audiotest.c is my sample program (it needs audio.h from the MusicPal kernel source code to compile) - it makes a continuous tone come out of the MusicPal speaker, which is progress! There are some comments in the code to show what it is doing (as far as I can work out anyway!).