Google neural network technology put to use in voice systems
updated 10:20 pm EDT, Sat October 6, 2012
Voice recognition first to benefit
Google will be using a neural network to work in its products. After creating software that can learn without the need for human guidance, Technology Review reports that the company will put its virtual brain "to work making Google's products smarter," with speech recognition to be the first to use it.
Earlier this summer, Google's X Lab created a billion-connection "neural network" that, after being fed 10 million 200x200 images from YouTube thumbnails, was able to recognize human faces and bodies over three days of learning. Although the accuracy rate was 15.8 percent, it was perceived to be a significant advance from before. It was also discovered that the same system was also able to recognize cats, often the subject of popular clips on YouTube.
According to Vincent Vanhoucke, a developer for speech recognition within Google, the system has received "between a 20 and 25 percent improvement in terms of words that are wrong." This means that there could be fewer errors in dictations and transcriptions in Google products in the future, such as Google Now potentially becoming a better listener. It has also been suggested that Project Glass could also benefit from such a network, with software capable of understanding real-world data more effectively, and responding appropriately.
