I worked on speech recognition (really, more NLU and question-answering, but close to the ASR team), and we have a few other LW members actively working in the area as well.
For careful (trained or at least practiced) speakers compared to touch-typing, on material that’s amenable to ASR (mostly not very precise and hard-to-predict symbols), speech can be faster by 1.5-2.5 times (80-120wpm typed, 120-200wpm spoken). But that’s not really the normal use case—many people find that for written output, they need the control and editing capability of keyboards more than pure speed.
Unless you’ve got a medical, safety, or convenience reason to use voice input (of which there are many!), typing is probably much better. The dream is a hybrid—switching seamlessly between spoken and typed input, depending on what’s best for any given unit of input. I haven’t seen any consumer-usable examples of this, though.
I worked on speech recognition (really, more NLU and question-answering, but close to the ASR team), and we have a few other LW members actively working in the area as well.
For careful (trained or at least practiced) speakers compared to touch-typing, on material that’s amenable to ASR (mostly not very precise and hard-to-predict symbols), speech can be faster by 1.5-2.5 times (80-120wpm typed, 120-200wpm spoken). But that’s not really the normal use case—many people find that for written output, they need the control and editing capability of keyboards more than pure speed.
Unless you’ve got a medical, safety, or convenience reason to use voice input (of which there are many!), typing is probably much better. The dream is a hybrid—switching seamlessly between spoken and typed input, depending on what’s best for any given unit of input. I haven’t seen any consumer-usable examples of this, though.