"Though I was amazed at how well it said some of my friends’ names (I’m Asian, so it’s not like I have a bunch of John Smiths as friends), saying “nees” instead of “nice” is unacceptable. Thankfully most words are pronounced correctly, but minor mistakes like these are noticeable."
His example got us to thinking about how Friends Aloud handles heteronyms -- words that are spelled the same but pronounced differently. The word "read" is the quintessential example, being pronounced differently according to present ("reed") vs. past ("red") tense. The word "nice," meaning pleasant, is pronounced with a long 'i' and "Nice," meaning the city in France, is pronounced with a short 'i'. When it's capitalized, such as at the start of a sentence (or perhaps improperly for emphasis as in "You are So Nice!"), it's difficult for the computer to know which pronunciation to use. So we started testing Friends Aloud on this very word in context, and we discovered that it actually does pretty well in this particular instance, leaving us to wonder just what the exact context of his mispronunciation was. For example, it verbalizes "Nice to see you" properly, even as it reads "We are going to Nice" correctly as well. It uses context to determine the pronunciation of heteronyms. Still, since Friends Aloud is not using full artificial intelligence natural language understanding, but only word context to figure out pronunciations, it is bound to get some things like this wrong. Just like a human reader would, frankly.
The problem here, most likely, was that the single exclamatory sentence "Nice!" was probably used in a post or comment and, not having any context at all surrounding that single word, Friends Aloud chose to pronounce it as the city in France. We will always keep working to teach Friends Aloud be less high-brow and to resolve such conflicts in favor of modern-day slang. In the meantime, rest assured that it does get at least most of this complicated language we call English correct, even when you use texting-style abbreviations.