There are technical reasons why tubes sound different from transistors, usually. Why there is so much controversy surrounding it is a mystery to me when there is plenty of objective reasoning to show they are different animals.
First, the tube is a higher impedance, higher voltage, lower current device than the transistor in almost all cases. This has to alter the design topology and interface between components when dealing with those differences alone.
Second, the triode is a unique device not duplicated in the transistor world in terms of its characteristic curves. There is a tiny but mostly unused and ignored portion of the curve of a FET (JFET, MOSFET, it doesn't matter) called the "triode" region. This is the only place it mimmics the behavior of a triode, and is basically so small an area as to be unusable. The FET is used in the "saturation" or "constant current" region, and mimmics more closely the pentode. A BJT is a current amplifying device that is also different than these others.
The triode has curves that have the following properties: nearly infinite DC impedance in common cathode configuration except for the biasing resistance, and current increases with increased plate voltage over the entire range of operation. Pentodes, like FETs, don't act that way because they go into constant current for increased plate (drain) voltage. In both cases I'm talking about holding the input voltage constant when raising the plate (drain) voltages.
This triode characteristic curve does an interesting service in a typical common cathode amplifer stage: as plate current increases from increased input voltage the plate voltage decreases as any inverting amp stage should, but with the decreasing plate voltage the tube wants to counter that increased plate current. This is local negative feedback built in without even trying.
So the triode typically has lower gain and lower distortion than a pentode or FET that does not have this local negative feedback working on its own.
You can make this triode amp sound more like a transistor amp by applying the same global negative feedback techniques that most of the transistor amps use, and you might like that. But there are limitations: triodes are lower gain and lower bandwidth usually so less NFB can be applied for similar overall gain. Thus you are left with some residual triode sound even if you try to squash its signature distortion.
It's not that there's a frequency response rolloff that accentuates the midrange. That midrange emphasis is more likely a feature of the distortion spectrum of the devices in the midrange compared to the extremes from two likely sources: A loose midbass with lowered NFB and poor PS regulation can produce harmonics into the midrange that decreases with increasing frequency. And then in the highs there is again a decrease in overall NFB which changes the uniformity of this distortion spectrum. The result is a frequency dependent distortion of unique triode-produced distortions, be it considered good or bad.
And while some people claim that distortions are not audible when "linear enough", also realize that in a bottle of wine that contains 14% alcohol and a lot of other complex chemicals, the whole bottle can be ruined by 5 parts per BILLION of the chemical that makes it taste "corked".
There are three kinds of distortions: audibly pleasant, audibly unpleasant, and inaudible. How much is audible of what kind of distortion is very difficult to find out through experimentation since it's almost impossible to alter just one kind of distortion at a time.
What is considered "most accurate" is still a judgment call. Many people say the triode amp sounds "most accurate" when compared overall to live, in most implementations, and I still agree with that. Why? I don't know and I can tell no one else has that down to a solid science yet.
Kurt
First, the tube is a higher impedance, higher voltage, lower current device than the transistor in almost all cases. This has to alter the design topology and interface between components when dealing with those differences alone.
Second, the triode is a unique device not duplicated in the transistor world in terms of its characteristic curves. There is a tiny but mostly unused and ignored portion of the curve of a FET (JFET, MOSFET, it doesn't matter) called the "triode" region. This is the only place it mimmics the behavior of a triode, and is basically so small an area as to be unusable. The FET is used in the "saturation" or "constant current" region, and mimmics more closely the pentode. A BJT is a current amplifying device that is also different than these others.
The triode has curves that have the following properties: nearly infinite DC impedance in common cathode configuration except for the biasing resistance, and current increases with increased plate voltage over the entire range of operation. Pentodes, like FETs, don't act that way because they go into constant current for increased plate (drain) voltage. In both cases I'm talking about holding the input voltage constant when raising the plate (drain) voltages.
This triode characteristic curve does an interesting service in a typical common cathode amplifer stage: as plate current increases from increased input voltage the plate voltage decreases as any inverting amp stage should, but with the decreasing plate voltage the tube wants to counter that increased plate current. This is local negative feedback built in without even trying.
So the triode typically has lower gain and lower distortion than a pentode or FET that does not have this local negative feedback working on its own.
You can make this triode amp sound more like a transistor amp by applying the same global negative feedback techniques that most of the transistor amps use, and you might like that. But there are limitations: triodes are lower gain and lower bandwidth usually so less NFB can be applied for similar overall gain. Thus you are left with some residual triode sound even if you try to squash its signature distortion.
It's not that there's a frequency response rolloff that accentuates the midrange. That midrange emphasis is more likely a feature of the distortion spectrum of the devices in the midrange compared to the extremes from two likely sources: A loose midbass with lowered NFB and poor PS regulation can produce harmonics into the midrange that decreases with increasing frequency. And then in the highs there is again a decrease in overall NFB which changes the uniformity of this distortion spectrum. The result is a frequency dependent distortion of unique triode-produced distortions, be it considered good or bad.
And while some people claim that distortions are not audible when "linear enough", also realize that in a bottle of wine that contains 14% alcohol and a lot of other complex chemicals, the whole bottle can be ruined by 5 parts per BILLION of the chemical that makes it taste "corked".
There are three kinds of distortions: audibly pleasant, audibly unpleasant, and inaudible. How much is audible of what kind of distortion is very difficult to find out through experimentation since it's almost impossible to alter just one kind of distortion at a time.
What is considered "most accurate" is still a judgment call. Many people say the triode amp sounds "most accurate" when compared overall to live, in most implementations, and I still agree with that. Why? I don't know and I can tell no one else has that down to a solid science yet.
Kurt