Humor me. When Gmail first came out, it was revolutionary. Now, you could host your own clone for $10-20/month using containerized Elastic Search indexing all of your email, with a small virtual machine ingesting your email from the internet, and providing it to you over JSON to your local mail client or through a web interface.
Voice recognition could follow the same trajectory. Computing power will increase, storage costs will continue to plummet (I've mentioned this multiple times on HN, but Samsung is predicting 128TB SSD drives in 3 years).
You don't need patents anymore, as you mentioned, because you can provide the service, not the underlying source code. But you still need to monetize quickly (if that's your goal), because the rising tide of technology can quickly supplant your lead with other competitors or open source options.
Pervasive Internet and the move to putting logic on the server definitely outmodes many uses for patents, though. Take Siri. SRI invented the technology, spun off a company with the IP, which was purchased by Apple. I don't think that transaction would've happened without patent protection. But today you have an alternative model. Instead of licensing voice recognition technology for integration into products, you can provide the algorithm as a service. Now, the source code stays safely on the server side where nobody can decompile it to figure out how it works.