Speaking of the hollering in the early records - that's another point for why the recording is fake. It's read quietly, intimately - and yet it got recorded. This would've been impossible before or in 1900, when people still had to speak very loudly for the sound to be picked up by the insensitive membranes. I read a technical analysis of the cylinder, and the conclusion was that it could've been made that early - the technology just wasn't around yet.
I might as well just copy-paste from the site
The recording as it sounded could not have been made in 1900 for a number of complex technical reasons:
1. The surface noise has a different frequency spectrum from the voice. The noise is bandwidth-limited to an octave or so centred on about 1kHz; the speech extends from about 200Hz to about 3kHz. While the latter is not absolutely impossible, it could only have been reproduced if the accompanying surface-noise had similar characteristics.
2. It is highly unlikely, however, that a frequency range as described above could have been achieved by 1900. In particular, the sibilants, though faint, are true, and do not have the characteristic diaphragm-resonance colouration endemic to all acoustically recorded speech, even Edison "Diamond" discs. Prior to the invention of the acoustically damped diaphragm (in Wente's microphone of 1924) this cleanness of sound could not have been achieved.
3. There is also no sign of horn colouration, though admittedly this could be rendered inaudible by a first-rate machine skilfully operated.
4. The speaker seems to be delivering the lines in too quiet a voice to have been recorded by the acoustic process. In all other surviving examples of acoustically recorded speech the voice is pitched as if the speaker were addressing a large hall full of people. This was essential to overcome the insensitivity of the recording diaphragm. The reciter of The Ballad of Reading Gaol, however, has adopted a very intimate tone by comparison.
5. The character of the surface noise is quite unlike that of other cylinders held by the NSA. It consists of discrete clicks (rather than hiss or swishing), which are characteristic of physical damage such as scratches or mould. There is very little rhythmic element to the sound, implying that such scratches or mould are always less than one groove-width (a hundredth of an inch) in size, because otherwise two or more neighbouring turns would be affected. Yet the clicks are loud, louder than the speech, indicating that the damage or mould is of quite significant dimensions. Also, there is no sign of the increase of noise at the ends of the cylinder which is generally part of such damage.
6. The rotational speed of a cylinder can usually be ascertained by listening to the surface noise. It is not easy to be dogmatic in this case because of the lack of repeated clicks, but the basic tempo indicates a speed well below the 120 rpm then considered standard--in fact, exactly as if someone had used a 78 rpm turntable to generate the clicks.
This analysis of the technical aspects enabled us to get beyond the existing debate and its reliance on memory to the internal evidence that the original recording could not have been a cylinder and could not have been recorded as early as 1900. Oscar Wilde died in that year: it could not be his voice on the recording.