5
$\begingroup$

It seems as though modern speech recognition (e.g., through Android, iOS phones) make use of grammar or sentence structure. (e.g., it might have a tough time distinguishing between "grammar" and "grandma" but can distinguish between "I'm going to see grandma" and "I'm reading a book on english grammar". (yes, I just tried it with my Android phone with vLingo app)

That is much improved (with Speaker Independent SR (i.e., no training)) over what I experienced with Dragon Dictate even using Speaker Dependent SR (with 30m of training).

So, I'm wondering whether my guess is right: When did the commercially available SR software start using grammar and sentence structure to "guess" the right speech?

$\endgroup$
5
$\begingroup$

As far as I know, much of the speech recognition software is build using Hidden Markov Models (HMM) where observations are individual sound phonemes, and the states correspond to words (or some partitions of the words).

I don't think the systems take into account grammar or sentence structure in the sense that they encode rules of English so much as they naturally pick up common structure from training. Because inferring the state in an HMM is based on both the observation (the sound) and the previous state, by adding more training sequences it is possible to capture typical phrases.

For example, there are likely several instances in the training data where the word "English" is followed by the word "grammar", whereas there are very few where it is followed by "grandma", thus the dynamic aspect of the model will capture these connections naturally.

$\endgroup$
1
  • $\begingroup$ I think you could view HMMs as equivalent to (stochastic) regular grammars. In practice, I hope they use HMMs like lexers but put a (stochastic) parser of some capacity on top (if they want to extract structure). A pure regular (also dubbed n-gram) approach is doomed (and known) to fail hard when applied to languages that have more complex grammars than English. Try translating German to English with Google translate... $\endgroup$ – Raphael Apr 6 '12 at 17:37
0
$\begingroup$

There is a pretty good description of the technology that Vlingo is using for speech rec here. They are using Hierarchical Language Models (HLMs) which does look at the context of words with in an utterance to achieve better accuracy. Alas, after being sued by Nuance for patent infringements, Vlingo has been acquired by Nuance. Soon Nuance will have a monopoly on all commercial speech recognition technology.

$\endgroup$
2
  • 1
    $\begingroup$ Welcome! What relation do your last two sentences have to the question? Also, are there (good) non-advertisement references to the technique they use? $\endgroup$ – Raphael Apr 6 '12 at 19:48
  • $\begingroup$ Just pointing out that every technology referenced in the question (Vlingo and Dragon Dictate) are now owned by Nuance who has a huge corner on the speech recognition market. If by "non-advertisement" you mean non-commercial there are some people who have been successful making solutions with Sphinx, which comes out of Carnegie Mellon University. I am in no way endorsing Nuance technology but just pointing out that most of the successful commercial solutions on the Android and IPhone (Siri) use Nuance. I also correctly pointed out that the tech used for better accuracy is HLM, not HMM. $\endgroup$ – Kevin Junghans Apr 9 '12 at 13:48

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.