I'm talking about Dragon Dictation [iTunes link] from Nuance, the developers of the very popular Dragon Naturally Speaking for the PC. Nuance also provides the speech recognition engine for MacSpeech Dictate on the Mac platform.
To dictate on the iPhone you just launch the app, press the record button, and start talking. Your dictation can be a brief sentence, or a much longer treatise. Once the text has been created from your speech, it's possible to email it, send it as a text message, or put the result in your clipboard. After recording your message, you can edit the resulting text before you send it off for others to read.
It's pretty slick! When you record your message, it is quickly transmitted to Nuance servers where a speech recognition algorithm is run against your data. The resulting text is returned to your iPhone very quickly; my informal benchmarks showed that it took about a second for text to be processed on a Wi-Fi network, and less than 5 seconds over 3G. You'll need a data connection for the app to work, but having this speech-to-text capability is going to be very important to a lot of people, who will find all sorts of uses for it.
I tested the app for about a week and found the accuracy to be very good. Accuracy diminishes if you are in a very noisy environment, as I found when I tried some dictation while being driven down the interstate. There were a few errors, but they were easy to correct. To add punctuation to your text, you can say 'period', 'question mark', or 'new paragraph,' and Dragon Dictation adds the appropriate punctuation.
Since your dictation goes out over the internet for processing, I asked Nuance about security. Their reply was as follows:
How much will this capability cost? Nothing. For a limited time Dragon Dictation is absolutely free, but at some point there will be a charge for it. I spoke to Michael Thompson, senior vice president and general manager for Nuance Mobile, and he said they see all sorts of specialized applications of this technology, such as in the legal and medical professions. The app has gone live and is available now."Search queries and dictation requests are transcribed by fully automated speech recognition software, without the use of humans. Data is uploaded and collected in order to improve performance for individual users, and to improve the general performance of the system.
All speech recognition requests and associated data are processed in data centers in the U.S. that meet stringent security and privacy standards; these are the same standards that we use for processing private information in other areas of our business."