Back in my days at Dialogic, I suggested a novel approach to a hard problem. Let's say, for example, that you want to do speech recognition of someone who speaks Hungarian. Here in the US, there isn't much call for Hungarian-language speech recognition, and no one is going to install it and maintain it. So how can you do speech recognition in odd languages, or avoid the cost of installing any speech recognition, even English-language, in the first place?
I suggested that we disaggregate speech recognition; if you're in the speech industry, you may have heard me suggest it at a few talks I gave. We'd collect the voice here in the US and make arrangements with people overseas to provide recognition in their language; when we wanted a Hungarian-language recognition done, we'd connect to their overseas system and perform the recognition there. We'd get highly-specialized recognition that'd be simple to use — the usual benefits of disaggregation.
We never implemented this at Dialogic, but I've just found a company that implemented this "recognition rental" business model. They charge 750 € to purchase a recognizer, but if you like you can send data to them for recognition for 0.05 € for each successful recognition.
Topics: · design · technology
Link to this story · Comment form · Blog Home
To leave a comment, please fill out this form.Comments are closed for this story.
Trackbacks are closed for this story.
