Singing synthesis research is surprisingly diverse; it is not limited to trying to make synth voices sound more “realistic” by any means. In addition, it has some surprising applications, as well.
For example, the article “Word Level Automatic Alignment of Music and Lyrics Using Vocal Synthesis” by Maddage and Sim details a system for aligning synthesized vocals with text lyrics, one that could possibly be used in the karaoke industry if it becomes well-refined enough.
Instead of using Hidden Markov Models, like most systems aligning vocals and text, the authors used synthesized vocal signals. They also used a TTS model of singing synthesis.
Now, this last point is the most interesting to me. Most commercial singing synth systems, if not all of them, use a system with a piano roll interface, and while, English vocaloids use some rudimentary TTS in the form of a dictionary, it makes me wonder what if somebody did develop a commercial singing synth that was closer to the TTS side of things. Would it be more or less usable than the current products out there? My guess is that it would be less usable, but I don’t know for sure.