3 ms·
That'd be interesting because it'd split the problem into "parse and highlight what should be emphasised" and "do the TTS".
by noir_lord 4y ago
That'd be interesting because it'd split the problem into "parse and highlight what should be emphasised" and "do the TTS".
- gattilorenz 4y agoI think there's alrrady research for "TTS after NLG" that does this, since a NLG system can export meta-info about emphasis, in addition to the text (at least in case of non-end2end NLG systems). Whether that makes a big difference in practice, I don't know.