Linked by Thom Holwerda on Fri 7th Oct 2011 20:48 UTC
Thread beginning with comment 492431
To view parent comment, click here.
To read all comments associated with this story, please click here.
To view parent comment, click here.
To read all comments associated with this story, please click here.
News
Linked by Thom Holwerda on 05/24/13 14:44 UTC
Linked by Thom Holwerda on 05/23/13 23:22 UTC
Linked by Thom Holwerda on 05/23/13 22:04 UTC
Linked by Thom Holwerda on 05/23/13 22:01 UTC
Linked by Thom Holwerda on 05/23/13 17:52 UTC
Linked by Thom Holwerda on 05/22/13 22:23 UTC
Linked by Thom Holwerda on 05/22/13 13:38 UTC
Linked by Thom Holwerda on 05/22/13 13:30 UTC, submitted by JRepin
Linked by Thom Holwerda on 05/21/13 22:06 UTC
Linked by Thom Holwerda on 05/21/13 21:45 UTC
More News »
Sponsored Links



Member since:
2010-03-08
Now, what are the problems which explain why computers took so much time to get this speech to test translation relatively right ?
First, there is the [word sliced in phonemes]->[written word] translation. It is not as simple as it looks, because many European languages have this "feature" that there are several ways to write a given phoneme. If you go in Asia, things are even worse : words are not commonly spelled using syllables, but using more complex characters which are often also words in their own right.
For all these reasons, voice recognition systems need an internal dictionary to associate a bunch of phonemes with a written word.
As a starting point, someone who wants to create such a dictionary can use a regular dictionary, take the phonetic expression of each written word, and create a phonetic-to-written dictionary from that. But if you stop at this stage, you'll miss all the everyday familiar vocabulary that is not officially recognized by national dictionaries, such as weasel words. These words, along with other things which are not found in dictionaries (such as the names of numbers, letters, and mathematical symbols) must be added manually.
Manually adding words that are not in the dictionary takes a lot of time and effort, and developers cannot think of everything, so some words will always end up missing. Especially taking into account that our vocabulary is in constant evolution. For this reason, good voice recognition systems must be able to learn new words. Which is a first form of AI.
Edited 2011-10-10 17:05 UTC