I was wandering past, and sensed the need for some serious clarification here...
Oversampling and upsampling are BOTH ways of changing the sample rate of a digital audio file (or stream) to a higher sample rate. When this is done inside the DAC itself (and usually in even multiples; ie 8x) it is usually called oversampling. When it's done somewhere else before the DAC (could be in a player, or a computer, or a separate box), it's usually called upsampling. Usually upsampling is done to some sample rate that is not an even multiple, but is instead a standard value (96k or 192k). Neither process is more accurate, or "better", and often both are used in one device.
Neither process, no matter how well it is done, can "create information", so the resulting digital audio CANNOT be more accurate than the original. The extra samples are interpolated from the original data, and, if the math is done well, they will not adversely affect the accuracy, but they cannot improve it. Oversampling does NOT mean "upsampling too much", and either process, if done correctly, is equally accurate. (Oversampling, to an even multiple, uses easier math, so is easier to do.)
So, then, why bother to do it?
The answer is simple. The highest frequency that a particular digital signal can contain is limited by the sample frequency (specifically, the limit is 1/2 the sample frequency); this is called the Nyquist Frequency. So, for a CD, with a sample rate of 44,100 , the highest frequency it can contain is 22,050 Hz (actually slightly lower). But, even more importantly, the conversion process results in all sorts of nasty noise and "byproducts" at frequencies above that 22,050 Hz. Without going into a lot of math.... you MUST use a high-cut filter to filter out EVERYTHING above that 22,050 in order to get back your original signal (and to prevent a lot of nasty noise and distortion).
Unfortunately, designing and building this filter can be a real problem. Audio extends up to 20 kHz, so we need a filter that passes everything up to 20 kHz without messing it up, but cuts off EVERYTHING above 20 kHz. Ideally, it should be down about 100 dB at 22 kHz. This is referred to as "a brick wall filter", and is impossible to actually make. In real life, you're stuck with a compromise that cuts off most of the stuff past 20 kHz, yet doesn't do too much damage to the audio band. [These filter compromises were why the early CD players often didn't sound very good.]
Now, let's try upsampling our signal to 192k. By upsampling, we have "magically" changed our filter requirement to one that is easy to implement. The audio information stays the same but, because we have increased the sample rate, the Nyquist frequency is much higher. Instead of needing a brick wall filter, now all we need is a filter that passes everything up to 20 kHz without messing it up (that part doesn't change), yet is down a lot at our NEW Nyquist frequency (96 kHz). This filter is a lot easier to design (it's actually possible and practical), and we can even build it with cheaper components and still get excellent results.
There you have it.....
The short answer is that upsampling and oversampling don't do anything to improve the audio quality; what they do is make it possible to design the (required) filter circuitry in such a way that it doesn't make a mess of the converted audio.... What they do is to make it possible for the DAC to do its job properly (which is virtually impossible to do without upsampling). And, finally, since most modern DACs do oversampling internally (it's referred to as an oversampling filter), upsampling outside the DAC as well is really more or less redundant.
