Imagine that you have a perfect recording medium that adds no noise whatsoever to any signal that you choose to record. Then imagine that you use this perfect medium to make a copy of an existing recording that has its noise floor at, say, 54 dB below full scale (the 0 level of the recording). Quick: What would be the noise level of the copy that you just made? That's right, it would be 54 dB below full scale.
The same is true if the recording medium is not perfect, but its noise floor is distinctly lower at all frequencies than the noise floor of the recording that you're copying. If the copy is being made on a system with a 95 dB noise floor and the original recording is as above, there will be no difference in the noise floor of the result--you might as well have made the copy with a perfect recording medium.
People, p-p-p-p-p-please let's get this straight: Word length in digital audio--the number of bits per sample--affects ONLY the dynamic range of the recording. You don't get any other improvements in sound quality whatsoever. More bits per sample won't give you "rounder" or less "stairstepped" signals, lower distortion, better "resolution," better sound quality at low levels (apart from the effect that noise always has, just like in analog media). More bits per sample simply makes the noise floor lower. So if you're recording a signal that has a lot more noise than the medium you're recording it onto, there's simply nothing to be gain by increasing the number of bits per sample far past the dynamic range of your source material.
It's 6.02 dB per bit, so you can easily do the math. If this is a very carefully made Dolby "B" cassette recording on pure metal tape (which was newly available in around 1974), 11 bits would be enough to render it. Of course you want the noise floor of the digital transfer to be several dB below the noise floor of the cassette all across the audio frequency range, so by all means go for 12 bits. But don't expect any sonic improvement beyond about 12 bits because by then the noise floor of the digital channel is much lower than the noise floor of the signal you're recording.
I'm sorry but all the references in this thread to "stairsteps" and such are simply wrong and the people who are saying them need to find out how digital audio recording actually works and stop giving advice until they have so learned. There, I've said it--I truly don't mean to antagonize any individuals in this very nice place, but some things are a matter of opinion while others are a matter of knowing how something actually works or doesn't work. Which you only prove that you don't know, if you speak of stairsteps. That metaphor outlived its usefulness in the 1980s. Die, metaphor, die.
--best regards