Google Now, Google's personal assistant like Apple's Siri and Microsoft's Cortana, may soon work offline and function seven times faster than current systems, a research paper submitted by Alphabet suggests. The paper talks mostly in a theoretical context but says the new system has already been tested on a Nexus 5.
At present, Google Now has very limited offline capability where advanced commands need to be sent and processed by a server. This causes delays and often fails to complete or comprehend commands due to unreliable networks.
To overcome this challenge, researchers at Google have come up with a number of complex computational and compression models with a local voice system that runs on average seven times faster than the current real-time speech recognition on a Nexus 5 and takes up just 20.3 MB of storage. To achieve this, the researchers used around 2,000 hours of anonymised Google voice search traffic, totalling more than 100 million utterances, while adding in noise from YouTube to imitate real-life speaking conditions.
The results, according to the paper, are based on two distinct domains, dictation and voice commands. In order to keep the disk space low the company has experimented with language model interpolation techniques that enable sharing of a single model across both domains.
The research paper will be presented by the company at the 41st International Conference on Acoustics, Speech and Signal Processing, commencing on 20 March.