Google lately affirmed that it’s developed a fresh offline speech recognition system that’s seven times “faster than real time.” The firm also shown that it examined the brand new system with no data connection on the Nexus 5, and published a research paper where it detailed the new offline speech recognition system.

The search giant’s outline for the released paper noted, “We describe a big vocabulary speech recognition system that’s precise, has low latency, and yet has a modest enough memory and computational footprint to run faster than realtime on a Nexus 5 Android smartphone.”

The business should never consume significant memory or computational resources and nonetheless included that it has to be precise.

The brand new system was realized by Google by including various elements in the general system design. “Using a mix of SVD-based compression and quantization, together with a streamlined first-pass decoding strategy and on the fly rescoring with a bigger LM (language model), we can construct a method which is about 20.3 MB in size, without compromising precision or latency,” described the paper.

For enhancing precision, Google also exposed its systems to sound and reverberation which was environmental records of day-to-day occasions and samples.

We understand that Google Now now has limited offline voice-command support. The listing of offline commands now accessible to users Open Gmail comprise Play Music, Turn on Wi Fi, Turn the volume up, Turn on the torch, Turn on airplane mode, Turn on Bluetooth, and Dim the display.

In other news, Google’s Senior fellow Jeff Dean in a recent interview affirmed that Google Translate will shortly have more precise results with heavy learning.

The VentureBeat points out that the deep learning system could be based on an 2014 printed paper labeled “Sequence to Sequence Learning with Neural Networks.”