By harvesting collective contributions from the masses, the smart get smarter. I'm sold on this approach, business model, science, whatever you want to call it... the mantra of cloud services like Search and other write applications where the wisdom of the group publishes larger knowledge than any individual expert contribution. Our contributions have built Wikipedia; our media built youtube, our keystrokes and abstract text donations made Google the most powerful learning machine ever, and now our voices too will likely add another dimension to Google's intelligence with the new Google Mobile App for iPhone (and planned support for other smartphones in development).
I've always admired Nuance's approach to speech input smarts where they learn from each user session and erect building blocks to master phonetic challenges. They're smart, and have grown smarter. I've also long been a big fan and user of Tellme and Vlingo.They're both smart and have grown smarter as well. That said, I can't help but think they are all seriously threatened or outmatched now, particularly if Google offers an API to other mobile developers once the service is perfected. That API, like all the others launched, would destroy licensing business models for voice recognition software and SaaS with one swift and precise blow.