Here's my theory: money. Whoever gets the first really good speech-to-text
product out there is going to make tons of money. Contrairiwise,
text-to-speech is clearly only going to make a few hundred kilos of money.
Hence, the smart money is investing in improving speech-to-text.
Some anecdotal evidence toward this point: I know three people who are
working on speech-to-text (and I know them by three different routes, so
it's not a circle of friends). I don't know *anyone* working on text-to-speech.
There may also be an element of 'the good is the enemy of the best': present
day text-to-speech is good enough to understand, so the motivation for
improving it is less than for speech-to-text, where present day tech is
still annoyingly inaccurate.
--CarlF