Speech to Text Coming To iPhone?

According to a patent filing, Apple is working on speech-to-text technology for its iPhone and iPod product lines. Speech recognition could be the holy grail for data entry and retrieval on mobile devices, especially as they continue to shrink in size.

The Baltimore Sun found the patent and has included a diagram of how the system would work when composing an email.

There is a lot of engineering speak in the filing, but I could decipher a few tidbits of info - that and I've seen this stuff on Star Trek so I know how it is supposed to work. It seems the speech recognition module they are working on would be able to not only handle text but non-speech data as well, such as punctuation.

To varying degrees, this has been tried before on mobile devices. The most rudimentary are the voice snippets you can record into your phone for a few of your favorite contacts. One of the better speech tools for phones is byMicrosoft (NSDQ: MSFT) and called Voice Command. It is really pattern recognition. You can say "Call Sally Jones at work" and it will search through your contacts and find a name that matches what your digitized voice said and dials the number. You don't have to train it or record her name before you can use it. You can also ask it the time, battery level, signal strength, upcoming appointments and more. It is rather limiting though and there is no way to compose an email with it or tell it to do anything outside of dozen or so tasks it was written for.

I recall one demo by Bill Gates a few years ago where he spoke into a Pocket PC (that is what they were called way back when) and got nearly flawless text recognition out of it, but the trick there was the voice data was converted to digital then sent via wireless to a powerful server which did the heavy lifting. It returned the text to the screen. In the day's GPRS networks, it just wasn't feasible, which is why he was using WiFi. Today with 3G networks, it is more realistic, but you have the issue of who is going to pay for the server to potentially service hundreds of thousands of voices simultaneously?

Read More Article...

Transcription Technology Watch

This is the first in a series of quarterly articles that will focus on technologies relevant to medical transcription. Hopefully, maybe even those MTs who are techno-phobic will find some of the topics enlightening, stimulating and/or of value in making career decisions. But maybe not. To challenge that hope, I've started off with everyone's favorite technology: speech recognition. If you want to really stimulate a transcriptionist, just say "speech recognition." Or, better yet, assert that "speech recognition will forever change the process of converting physicians' thoughts and utterances into text." Then run for cover.

Every transcriptionist out there has heard some form of that assertion. Their reactions range from dismissal to fear to anger. So what's the truth? What does the future hold? Well, at some point in the future, there will be no medical transcription. Physicians will dictate into a PC or portable device; their speech will be converted to text; and the dictator will make any necessary corrections to finalize the report. No transcription expense. No transcription delay. But that future is at least 3 years off. Just kidding. It's way more than that. However, there is a future closer than that, related to speech recognition, which has some major implications for this industry.

Doctors hate doing anything that they believe is below their stature or slows down their ability to generate revenue. So we will not see "front-end" recognition-where they correct their own mistakes as described above-in most environments for many years. But there's a new game in town. It's called "back-end" speech recognition. Physicians don't change a thing in their dictation behavior. They continue babbling into telephones or some other dictation device just like they always have. But their voice files are now run through a server-based recognition engine, a draft is produced, and a medical editor corrects the errors both in recognition and dictation.

Read More Article...