Even if it’s not one of the most used features, because not all Android users are English native speakers, Voice Search has an important role and the Google Speech Team improved the system in order to be able to identify words faster, and not only fragments of sounds. Now, Google App’s Voice Search will be more efficient in noisy environments, but more about this feature and its changes, you’ll find out in this article.
On the Google Research Blog, the company announced that they’ve added new computational models and they are “are more accurate, robust to noise, and faster to respond to voice search queries models.” This will make the voice search feature 300 milliseconds faster, but in some cases, the application will feel even 500 milliseconds faster, if the device has high end specs.
Google gave more details about the new application and how accurate it is, saying that the “neural network acoustic models” have been improved by using Connectionist Temporal Classification and sequence discriminative training techniques. We didn’t get a metric on the accuracy improvements, because Google didn’t go into detail too much, saying only that these are extensions to the recurrent neural networks, helping the voice search to be more efficient in noisy environments.
If we go back in time, three years more exactly, we remember that the Gaussian Mixture Model (GMM) has been replaced with Deep Neural Networks (DNNs), finally, after 30 years, and this new standard provided better results for the speech recognition system, no matter what the users were saying. Now, Google has replaced DNNs with Connectionist Temporal Classification (CTC) and this technology relies on rely on Recurrent Neural Networks (RNN), which “have feedback loops in their topology, allowing them to model temporal dependencies: when the user speaks /u/ in the previous example, their articulatory apparatus is coming from a /j/ sound and from an /m/ sound before.”
For words such as “museum”, the technology uses a Long Short-Term Memory (LSTM) RNN, which is able to memorize information better than RNNs because of the advanced filtering mechanism (memory cells and a sophisticated gating mechanism). Google didn’t stop here, and they pushed things further, training the models to recognize phonemes without making predictions every instant, and now they output a sequence of “spikes”, revealing sequence of sounds in the waveform. Thanks to the artificial noise which trains the sequences, the words are easily recognized in noisy environments. The 300 milliseconds delay, which affected the way the model recognized the phonemes, has been fixed, and the model has been trained to predict the phonemes faster.
Google integrated these models into the Google app for Android and iOS, which the dictation with the new model is live on Android OS, but Chrome OS the Chrome desktop will have to wait until these changes will be available for them, as well.
This application will help you find information on the internet or on your device, and to search for a word or phrase, you will use your voice. To “wake” the voice search feature, you will say “Ok, Google” and after giving the voice command – let’s say that you’re looking for a restaurant or hotel, this feature will provide you results based on your location (this means that it will also collaborate with the GPS enabled on your device). You can order this application to give you sports scores, traffic alerts, updates on movies, events, weather or news. More than that, “Ok Google” can be set to remind you to make calls or to lock your door.
The information you’ll require will be delivered on cards and you’ll get to choose the ones that are more relevant. On the google.com/landing/now/#cards website you’ll find examples of customized cards, but you will need to sign in, in order to get them.
How To Turn on “Ok Google”
When saying “Ok, Google”, you will “inform” Google app that your will use your voice to give commands. However, in order to turn on “Ok, Google”, first, you will need to get the latest version of the Google app from the Play Store, then open it and head to Menu icon menu icon from the top left corner of the page. Tap on it, then access Settings>Voice>Ok Google and check the box next to “From Google app.”
Other devices have a different method to turn on Ok Google, outside the Google app, by checking the box next to “From any screen” / “Always on.”
The settings you can change, after turning on “Ok Google”:
– “From any screen” will allow you to say “Ok Google” when the device is charging or the screen is on;
– “Always-on” works with a Nexus 6, Nexus 9 or Samsung Note 4, allowing you to say “Ok Google” even if the screen is on or off;
– “Trusted voice” will recognize the sound of your voice when saying “OK Google” from a secure lock screen;
– “Personal results” will provide results even if the device is locked etc.