I believe many articles detailed Siri, the difference between Cortana and Google Now, overall, Google is Now currently on the market is the most mature voice Assistant. Now, in order to strive for excellence, Google unveiled updated versions of its official search sector.
Existing voice model relies heavily on LSTM (Long-Short Term Memory), artificial neural network as a long term memory, it can be seen as RNN (recurrent neural networks) is a special extension. Compared with the rest of RNN, LSTM in the memory unit has unique advantages and gating mechanisms can remember more information.
(RNN structure)
Prior to this, Google Now uses DNN technology, which is currently the most popular technologies in the market, and was dominating before the whole industry for more than 30 years of GMM (the Gaussian Mixture Model) technology.
Before the waves dying on the beach, and the younger generation.
Now, rely on more mature LSTM technology, Google Now speech recognition faster and more precise, even in a noisy environment, also have good performance. How to eye movement routes to design Web pages
In addition, phoneme forecast (phoneme prediction) is 30 Ms delay this, Google search specifically to carry out the training, making it closer to the actual speed feedback loop function you can absorb in an instant you say all the words and voices, even though some sounds you haven't fully come, it can use the word "go".
(feedback loop)
--In other words, you haven't finished, it already know what you want to say.
Have added this service currently on the Google app, Android or iOS users can use.
No comments:
Post a Comment