I've enabled it only for 3 podcasts since these were the most accurate for analysis. It turns out the analysis works better when the speakers are not speaking or laughing over each other, otherwise the script can't pick up the cues that would indicate the end of a sentence. If you know any more podcasts like that, I can analyze them and add them to the list.
The next step will be to analyze sub-phrases within sentences and add an extra level of splitting, particularly for long sentences. E.g. it might be nice in Korean to add a sub-sentence split after suffixes like "면" and "하고" in longer sentences.
Then, I'd like to do something where if a long sentence has two parts A-B, the player could play it like this:
A
A
B
B
AB
AB
i.e. first repeat each part, then listen to the combined sentence. This is a simpler version of what I was doing in my listening attention span stretching experiment, although maybe I could build the full listening algorithm into it later.
Also, I've already mentioned the analysis is not 100% accurate, and it's made a little worse due to the fact that jumping to exact timestamps in a web player cannot be perfectly exact, so I've added the controls to add padding on the left and right side of each sentence. I tried the same exact sentence detection algorithm on my experimental mobile app, and the accuracy was spot on there, so it's just the nature of the web, although Firefox turns out to be much more accurate than Chrome when I tried it.
If you have any suggestions, please let me know. Or if you notice any long sentences that you think should have been split, I'd be interested to know where you think it should be split and why, then maybe I can come up with an algorithm that splits it this way.
I'm aware on occasion that sometimes a sentence boundary will go undetected, and you'll hear two sentences connected together. Sometimes there's nothing I can do about this due to the approximate nature of the analysis, but I think once I add sub-sentence splitting, maybe this can double as a failsafe against accidentally missing a sentence boundary. i.e. if the sentence boundary is missed, hopefully it will still be picked up as a sub-phrase split.
Oh, also, you may notice some quirks related to the fact that this is working with a remote mp3 file over the web. When jumping to a timestamp, it can sometimes take time to jump there because it may have to make a new request to the server hosting the mp3. If the mp3 has sufficiently buffered, you won't have that problem.