Hi,
I love using the new natural voices for reading aloud PDF books. However, there are a few issues with this feature currently:
- Selecting text in the middle of a large book and using Read aloud from here feature for the first time after opening a book results in a quite long delay (can be more than a minute) before it starts reading.
- When using reading speeds above normal text highlighting feature of currently spoken words starts to lag behind.
- Text formatting and structure (like line breaks, superscript annotations etc) can cause the reader to misinterpret how text should be read; resulting in artificial pauses and incorrect flow of speaking. I know that this one is inherit to PDF document format itself and asking for solution is probably a bit far fetched, but with the recent rise of machine learning, maybe with good data set of PDF books and source documents, a neural network could be trained to better interpret the flow of PDF documents.
Tested on Canary Edge version 86.0.587.0 and on Dev Edge version 85.0.564.13.
