I was thinking about this the other day. Recordings have a potentially much higher quality than a speech engine. There a lot of reasons for this, but a couple

I was thinking about this the other day. Recordings have a potentially much higher quality than a speech engine. There a lot of reasons for this, but a couple of them, like mispronouncing words, or the wrong cadence, are fixable through human intervention in the engines interpretation of the text. Wouldn't it be interesting if we had a speech engine that could take hints that were annotated in the text? Similar to how one can add type hints to clojure variables to clue the compiler in for better optimization. Perhaps it could even be unobtrusive and even readable, a la markdown, such as a tick mark to indicate a stressed syllable or earmuffs (asterisks) for an emphasized word. Perhaps inverted angle brackets >>could indicate speeding up a passage.<< With such a system, it would be possible to use something like git/github to take "patches" for improving the speech of a piece of text. This came to me when I was thinking about reading some of PG's essays to create audio versions that would be useful for long drives and accessibility. I emailed him asking for permission, but no response (I didn't expect one, he probably is flooded with email and/or didn't have the time for a response, but I thought I would try anyway.) I suppose I could just start doing it, and if it garnered attention and ran afoul of YC's sensibilities, I'd get a response then. However, it occurred to me that 1) my voice isn't particularly pleasant, and 2) audio files are very inefficient storage of textual information. Text to speech, while greatly improved over the decades, still isn't much of a substitute. But perhaps with hinting it could be good enough.