Demonic_Duck
Member
I recently acquired the "text to speech" function, and whilst I like it rather a lot in a "shiny new toy" kind of way, at the moment it does seem more than a little buggy. Sometimes it will read out words in a way that sounds unnatural, or worse, simply incorrect. At times this is due to 多音字. For example, "大率" is read as "dàlǜ" rather than "dàshuài", even though it's being read out from a dictionary entry (规范) which clearly indicates the pinyin as being the latter.
On the other hand, sometimes the misreading seems to be because of a problem with some of the recorded syllables themselves. For example, "免疫系统" sounds more like "miànyǐ xìtǒng" than "miǎnyì xìtǒng". However, "免疫反应" sounds more like "miànyí fǎnyìng". In other words, there's a clear difference in the way "免疫" is being read out (though in both cases it's incorrect). I don't understand very well how text-to-speech functions actually work, but clearly there's more at work here than simply mapping a single character to a single pronunciation and spitting them out one after another.
I guess my question is, are these the type of bugs that can be easily worked out or not? It seems like at least the problem with 多音字 in headwords, for dictionaries containing pronunciation information, should be fixable in theory.
On the other hand, sometimes the misreading seems to be because of a problem with some of the recorded syllables themselves. For example, "免疫系统" sounds more like "miànyǐ xìtǒng" than "miǎnyì xìtǒng". However, "免疫反应" sounds more like "miànyí fǎnyìng". In other words, there's a clear difference in the way "免疫" is being read out (though in both cases it's incorrect). I don't understand very well how text-to-speech functions actually work, but clearly there's more at work here than simply mapping a single character to a single pronunciation and spitting them out one after another.
I guess my question is, are these the type of bugs that can be easily worked out or not? It seems like at least the problem with 多音字 in headwords, for dictionaries containing pronunciation information, should be fixable in theory.
