We cannot even agree on what wide and narrow even mean with contemporary radios - are we talking 'old' wide, as in the old 25KHz deviation level, where 'narrow' was the reduction in bandwidth for 12.5KHz channels, while now narrow is even narrower. Programming new radios is always worth a check - some programmers will be setting 'narrow' because they've always set narrow, and I did it too, until I realised that Kenwood 'wide' in analogue is a much better match for some Icom 'narrow' radios, just a few years older!
I use both walkie talkie style radios and In-ear Monitor systems, and the difference in quality is huge. The IEMs, designed for audio quality are pretty transparent - with ear moulds transducers that cost the same as the receiver, quality can be very high - but at this level, the commanders fitted to maximise dynamic range need to track well together, or the audio sounds wrong - over compressed or worse, over dynamic. All this with a 20-15K audio bandwidth. Moving to comms, and because it's essentially speech, the bottom below 300Hz isn't just reduced, it's removed so the tone signalling works properly - with some CTCSS tones over 200Hz! Again - there's very little useful information above 4K, and Bell discovered this a very long time ago, so it can be rolled off - not removed this time, just gently filtered off. Move to digital and new problems jump up. The audio brigade continually slag off compressed formats like mp3 - because it's throwing away data, yet to most people - it still sounds good. You can select how many times a second you sample the audio, and how accurate in terms of different steps of loudness you make the sampling. The common goal is to sample at twice the highest frequency as a good compromise - this is still quite a high data rate to send as RF in a limited bandwidth maintaining a low error rate end to end, so the designers can tweak that cut off to the highest frequency to manage the data stream. With limited resolution in bit depth, they also need to compress the audio pretty severely, digitise it, transmit it, then reconstitute it and that's a very large amount of wrangling to get it back. You can see how a manufacturer simply designs their entire processing train to be their version of 'best'. If they wish to increase intelligibility and add crispness and reduce the robotic sound, they can do it - BUT - at a cost. Something else has to go, or suffer. They all pick a different point on the good to bad scale, and it gets worse when you mix design philosophies. The SDR systems that decode lots of digital formats cannot do it properly - they have no way of telling if it's a Motorola or something else running Motorola's version of the protocol. Mismatching cut off frequencies, filter curves, companding ratios and processing means they really don't play well between models, let alone brands or a specification. Sometimes mismatches work for you - usually it will be the expansion process not matching the compression at the transmitter. Over expanding can give what appears to be cleaner speech, at the expense of maybe a little extra distortion - an error the other way softens the response making it appear muffled or difficult to understand. It was simple in the old days, when was that? Last year?