Last night, Lissa Kay asks if I know how Jott works. I used to work at The Learning Company in its foreign language division which was bleeding edge when it came to speech recognition (which is different than voice recognition btw). I knew all about Lernout & Hauspie and how to trick Dragon Speech. Jott’s accuracy (not demonstrated last night) has always amazed me because there is no training involved. You sign up for an account and instantly start using it. Most speech-to-text software requires some training which usually involves reading several paragraphs of text to the software so it can learn the nuances of your speech patterns.
As it turns out, Jott combines machine translation and humans to convert the speech to text.
If we were dealing with a very limited set of words, in a known context, spoken very clearly by a accentless person in a noise-free environment, then pure machine-driven Speech Recognition might have been the way to go. Instead, we wanted to be immediately useful and simple to adopt, letting any English speaker jott using an ordinary cell-phone, their natural voice, in a realistic setting (their car, running between meetings, etc.). So we use a mixed Human/Machine method for transcription, and that blend will change over time. [Source, Entrepreneur27, Interview with John Pollard of Jott]
Since Jott uses humans, you can spell difficult words to assure they get turned to text correctly. I have inquired to see if Jott plans to support IM, SMS, and emails sent to the inbox.
See a screenshot, the numerous services to which Jott can post, and my thoughts on Jott at the bottom of Rough Week Behind Redux to Follow.
