
W. Sidney Allen wrote Vox Graeca[1] and Vox Latina which are well-known works which detail both what we know about pronunciation as well as explaining how we came to know these things. We surely don't know everything and scholarship evolves, but it is hardly like what you seem to be suggesting.
It is interesting to know about some of the sources: ancient grammarians wrote about pronunciation, people poked fun at how other people talked in letters, epigraphers sometimes made mistakes on the order of "could of".
[1] https://www.cambridge.org/us/universitypress/subjects/classi...
[1]: https://blogicarian.blogspot.com/2012/10/ancient-greek-food-...
I truly understand that the ancient pronunciation sounds bad to your ears! All those "mistakes" that you mentioned are exactly the ways that Greek pronunciation has changed over the thousands (!) of years. If you look closely, it is as if the writing system is made for the old language, which is the case. The modern pronunciation drifts further apart from what is written. Historical linguistics is quite interesting once you get into it.
I recommend the following books, although both quite technical in the linguistic sense:
Allen, Vox Graeca: The Pronunciation of Classical Greek 3rd Edition (Cambridge University Press 1987)
Horrocks, Greek: A History of the Language and its Speakers, 2nd Edition (Wiley-Blackwell 2010)
Classics:
Kafka's The Process and America - both stunning novels that deserve to be read as much as his most famous, The Castle.
Jan Potocki, The Manuscript Found in Saragossa - a trippy collection of tales organised in multiple nested frames.
Scott Fitzgerald, The Great Gatsby.
Leo Tolstoy, Anna Karenina. Still as breathtaking as I'd remembered it from the previous reading five years ago.
Science Fiction:
Gene Wolfe, The Book of the New Sun. Wolfe, whom I hadn't read before, turns out to be in a class of his own. It's literary SF, overwhelmingly masterful and beautiful in its prose and characters, yet it could give any hardcore SF novel a run in terms of its ideas. It demands and richly rewards a close reading. I'll be reading more Wolfe in 2010.
Ted Chiang, collected stories. Chiang writes only stories, and has written just a few of them, but nearly every one is a gem. Read Understand on the web (http://www.infinityplus.co.uk/stories/under.htm) to see if you like his style - I do.
Vernor Vinge, A Deepness in the Sky, The Peace War, Marooned in Realtime. All great novels, but start from A Fire Upon the Deep (I read that in 2008) if you haven't read Vinge before, it's the best introduction.
Greg Egan, Diaspora, Quarantine, Distress. Egan is the best hardcore SF author out there. _Diaspora_ is a fascinating take on shared simulated reality that's very different from the usual fare. Quarantine will appeal to your inner quantum mechanical geek.
Contemporary fiction:
Pynchon, V. Pynchon is the best novelist we've got. I recommend The Crying of the Lot 49 as the best Pynchon to start from; V is more difficult and much longer, but rewards the patience. I still can't get over the fact that Pynchon wrote and published this novel when he was 26.
Orhan Pamuk, The Black Book.
Annie Proulx, Fine Just The Way It Is. A new collection of stories from the best writer about rural America out there. Three are extraordinary, two so-so, the rest very good.
Non-fiction:
W. Sidney Allen, Vox Graeca and Vox Latina. If you want to know how ancient Greek and Latin sounded, these are the books to read.
Elizabeth Lloyd. The Case Of The Female Orgasm. An interesting study into how available evolutionary explanations of female orgasm fail in various ways. This is a book about evolutionary biology and its methodological soundness as much as it is about explaining female orgasm.
Computer-related:
Peter Seibel, Coders at Work. A fantastic collection of interview with great programmers. The one 2009 book no hacker should be without.
Squeak by Example. Does a great job of explaining Smalltalk in general and Squeak in particular.
W. Sidney Allen's old Vox Latina https://www.cambridge.org/core/books/vox-latina/0D460CEF06E5... is apparently still the standard starting reference for classical Latin pronunciation, at least for English-speakers. (Many nineteenth-century German philologists died to bring us this information, of course.) People such as Luke Ranieri on YouTube use a version of Allen's system, though a number of people including Ranieri claim that there should be five vowel qualities rather than the seven described by Allen https://www.youtube.com/watch?v=eH8E5RKq31I .
Note that if you want to speak Latin that's roughly faithful to how it was spoken up to (but not necessarily including) late antiquity, pronunciation is actually less important than quantity: basically, using clearly distinguishable short and long vowels in the right places (plus not running together double consonants in some places). I suppose it's similarly important to get the stress right, but at least that's generally agreed to be pretty easy. Classical Latin quantity feels weird and unnatural to English-speakers, and to Romance-language speakers, German-speakers ... : words often include one or two or three unstressed long vowels before getting to the stressed syllable, which might or might not itself have a long vowel. Even people who advocate for (classically-)correct quantity often don't consistently get it right.
(And yes, Allen also did publish a Vox Graeca https://www.cambridge.org/ie/universitypress/subjects/classi... , too, but be careful: the pronunciation of Ancient Greek is a question that might actually get you into a fistfight https://www.youtube.com/watch?v=BybLbHPU7Qc&list=PLQQL5IeNgc... .)
