Google’s researchers looked at how they could embed a “large vocabulary speech recognition system”, which is accurate, has low latency, doesn’t use up too much memory power on a Nexus 5 phone.
Google Now or the AI voice-command feature that we’ve all come to use on our Android smartphones, could work even in offline mode, according to a research paper put out by a team of Google Researchers. Essentially Google’s researchers looked at how they could embed a “large vocabulary speech recognition system”, which is accurate, has low latency, doesn’t use up too much memory power on a Nexus 5 phone.
The paper notes that currently running voice-commands and dictations require constant and fast internet connectivity, something that might soon be a thing of the past if the new system is implemented.
The paper notes, “However, despite increases in speed and the availability of mobile internet, speech recognition requests frequently have high latency, or even completely fail, due to unreliable or unavailable network connections. An embedded speech recognition system that runs locally on a mobile device is more reliable and can have lower latency; however, it must be accurate and must not consume significant memory or computational resources.”
As ZDNET points out that the idea with the paper is to create a voice-recognition system that runs locally on the smartphone. The researchers decided to integrate the contact lists within the system to enable efficiency for commands like ‘Send an email to X or Y.”
Read the full paper here.