Improvements have been made, and now Google Speech Recognition understands what you’re saying much better than before, with an upgraded system that will make speaking instead of typing much simpler. It’s a well known issue that voice recognition is not perfect.
- Google Now has received an update on its services
- The software has changed from deep neural network (DNN) to recurrent neural network (RNN)
- It will now better understand the difference between similar-sounding words
- Its predictions will take context into consideration
Google is taking steps to improve on the reality that there will always be discrepancies between what you say and what a device understands, otherwise we would all be talking to our phones on a constant instead of typing it out. Google Now is one of the voice recognizing software out there that has seen an important update.
The company has recently stated that they have made significant improvements to the acoustic models on their voice recognition app of Google Now, for both iOS and Android devices. Users will have a faster, better and more accurate way of using the search engine without typing anything with their fingers.
According to the company, the new type of recurrent neural network (RNN) will work far better at memorizing and interpreting information sent to your phone by voice, than the deep neural network (DNN). They have enhanced the way the software recognizes the difference between similarly-sounding words and pronunciations.
The statement was featured with many technically detailed statements, but the gist is that they have invested more training in their acoustic models to understand your voice, even through noisy surroundings. Which is the unfortunate situations most of us find ourselves in when we need Google Now the most.
The team of researchers added artificial noise while they were training the data in order to better help the software understand how to sort through the background sounds and focus more on the voice.
They also added improvements on latency and predictions, essentially training the acoustic models to take in data in larger chunks, which greatly improved on its speed.
Its prediction models have seen an update to adds to the software’s power to listen ahead of the speech signal, essentially making a decent connection between words and their meaning.
It’s reportedly a huge enhancement on previous models that is set to make the feature much more useful to users. And most who had attempted to conduct any activity with speech recognition have bumped into several flaws that deemed the program nearly unusable. Longer words proved themselves to be a pain.
Google’s voice recognition now actually takes context into consideration, using its prediction feature while carefully paying attention to its surrounding words.
Image source: gizmodo.com