As a software engineer with a strong hardware background, I can confirm that the digital source is pretty much irrelevant. If there is any relevance at all, it applies only to the medium.
The digital data stream, once it is on a wire, is what it is. There is no such thing as a string of bits that is "better". Where problems can arise is when the stream comes from something that is essentially mechanical, such as a compact disc. The readout of the data on the disc may not be perfect. There may be dirt, scratches, and so on, causing the readout of the bit stream to produce bad data. There are a whole bunch of tricks that allow a CD player to compensate for this. In particular, there is a decent amount of redundancy built into the encoding, so errors in the stream can be detected and, in many cases, 100% corrected such that the original data remains intact, even though there were some errors during reading.
Some errors are too serious to be repaired. In that case, a chunk of data in the stream goes missing. Good DACs will compensate for this by interpolating the missing data. In essence, they replace the missing piece of data by making a guess as to what it might have originally looked like by looking at the intact section of data just before and just after the error. In many cases, that is good enough to mask the error such that no human will be able to detect it. If the section of missing data gets too large, no amount of guessing can fix it, and you get a drop-out, where there is a moment of silence, or the interpolation is asked to do more than is reasonable and produces a bad guess, that is, distortion.
The important point here is that, in no case is there ever any doubt as to whether an error did occur or not. The DAC can tell with 100% accuracy when there is an error (any error, no matter how small or large). A DAC will never operate on a data stream that contains errors without knowing that there were errors.
If we take the mechanical component out of the equation, the picture changes. For example, if the source data is a Flac encoded stream that is read back from some storage device, such as flash memory, for all intents and purposes, the bit stream that comes out of the memory is 100% error free. The same applies to transmission mediums, such as wires, ethernet, optical fibre, whatever. Modern digital hardware is so good that it effectively never corrupts anything ever.
If we got the Flac stream onto the medium without errors, as is the case with direct-to-storage recordings, the bit stream that ends up on the storage device is also free of errors. Simply put, modern hardware makes no mistakes; at least not by any reasonable standard. I'd be far more worried about getting hit by an asteroid than some bit-level error in the transmission chain.
The only other sources of errors are on the analog side, say with the microphone, the microphone amp, or the analog-to-digital converter that turns the data into digital form. But those are things that, as a listener, I can't do anything about. Whatever distortion was produced during recording will be faithfully encoded in the digital data.
The other error source is in the playback chain, when the data is converted back into analog and processed from there. Lots of potential for errors on that path, in the DAC, in the amplifiers, etc.
But improving the quality of the digital source is likely to be a lost labour of love. There is no such thing as a "slightly better stream". The stream is either correct or it is not; there really is no in between. And because digital hardware is as good as it is, the stream is effectively always correct.
I am aware that no memory is perfect, that there are observable errors in hardware, etc. But they are too rare to matter in this context. My chance of dying in a car accident today is millions of times higher than me encountering a bit error in some digital data stream in the next twelve months.
