19 March 2019
Google has been one of the front-runners when it comes to working on AI-related technologies. One such offering from the company is speech-to-text functionality available on almost every Android smartphone. It has been there for years and Google keeps trying to make it fast and better. In the same direction, it has made significant progress which will undoubtedly benefit users.
With this progress, Google has announced what it describes as an "end-to-end, all-neural, on-device speech recognizer". It is currently available on Gboard but only for Pixel phone users. It will be offering a faster speech transcription but doing all the complex process locally on the smartphone itself. It is huge progress considering all other technologies and companies do so on remotely situated cloud servers.
The local processing means that the whole thing will happen a lot faster as there will not be a need for sending data to servers and then receiving transcripted data. It also means that users will be able to use text to speech functionality even when offline for the same reason explained above. Another noteworthy aspect is that it will do the transcription character-by-character instead of the usual trend of word-by-word. It means it will feel just like typing out something.
Now you might be wondering all of that is good but would not it take a lot of space on my phone? It would have been true for previously used models which weighed 2GB and were later reduced to 450MB. But with the newly announced speech recognizer, the models are only 85MB in size which means you should be able to store them on any Android device, even the ones with 8GB of internal storage.
Google says that the new speech recognizer currently works only for American English and on Pixel phones. But like any other technology, it will be working on expanding it to more languages and devices.