Google is researching ways that will enable its Google Now assistant to work independent of an Internet connection. Till now, voice commands on the Android platform have required an Internet connection but Google is working on trying to create a native language processor that allows for the same capabilities without taking up too much space on Android phones.
The possibility of an artificially intelligent assistant on Android has already been realized to some extent but all the computations are outsourced to Google’s servers, which means that an Internet connection, whether it be Wi-Fi or phone data, is a necessity to use it. Running the commands through a server also means that it takes longer for the whole computational process to take place. Google is reportedly tackling the issue and will soon be publishing papers regarding its attempts to make the commands work without a connection.
Google’s attempt to bring speech recognition to Android machines will bring new levels of personalization to users; not to mention greater responsiveness. Google also plans to keep the technology very conservative in terms of memory, reportedly just around 20.3 MB. A minuscule amount and quite an accomplishment if you consider that Google’s speech recognition system has 2,000 hours of Google voice records from a 100 million unique commands and has been taught how to cut the noise out of these. To fit it into the small 20.3 MB package, complex compression models must have been applied and results of the machine learning have been converted into complex computational models which allow the Android devices to understand their user’s speech.
Android enthusiasts will be awaiting the 41st International Conference on Acoustics, Speech and Signal Processing on March 20, to learn more about where Google has gotten and how far we are from a full AI on our little dumb Androids.