I've explained this many times before. I don't remember where and I don't feel like finding it, but there's a really good explanation out there somewhere. Probably in either the trusound or truvid threads.

The basic idea is that you need to use pulse width modulation to approximate different voltages. You're going to be flipping the link port 18,000 times per second. This means you're going to send 18,000 groups of ON OFF. Now, the timing of your ON OFF determines the approximate voltage. 50% ON 50% OFF is of course the middle voltage. 0 100, 33 67, 67 33, 100 0 gets you 2 bits of output with 4 possible positions. The more voltages you have the better it's going to sound obviously. I think in trusound I used something like 250 steps.

To finally play your music, just run through the data and output the correctly timed ON and OFF pulses. The main downside to this is that depending on how you implement it, it takes 30%-100% of the cpu power just to play music.

You can use pwm or you could use external hardware like a digital to analog converter with a 2 wire interface.

You have to see with the remaining program space you can use and the available processing power to determine at which rate you can play the audio. I honestly don't believe you will be able to play audio at 18 khZ, but this seems to be a starting point. You can lower the quality after if it slows down your program.

If I looked at your timing loop well, once it is done rendering a frame it waits for your 1/30th second to render the next. You will have to change your timing loop for it to process your waveform audio before waiting for the next frame. Basically here your goal will be to calculate your image and audio generation loops to have as few waiting time as possible between frames.

So for PWM it will really be 36,000 writes to the link port per second? And if I am doing PWM at 18KHz, won't the noise from that be audible?

Yes and yes. I would actually recommend using two crystal timer interrupts to do the audio output, the first repeating at the sample rate and the second scheduled by the first. As for the noise problem... I'm not sure exactly what you should do about that. I guess increasing the sample rate would be out of the question. (Well actually, you could perhaps reduce the audio quality to 6-bit or something, there's really a limit as to how accurate your PWM timing can be anyway)

By using a simple audio filter you can filter out anything above a certain frequency.