Google has announced that due to “our new acoustic models” now used for voice searches & commands in the Google app (on Android & iOS), & for dictation on Android devices, voice search had become even more faster & even more accurate.
Announcing this from its official Google Research blog, the “Speech Team” wrote that in addition to requiring much lower computational resources, the new search was also more robust to noise, & faster to respond to voice search queries.
It was in 2012 that Google voice search had adopted Deep Neural Networks (DNNs) as the core technology used to model the sounds of a language. These had replaced the 30-year old standard in the industry: the Gaussian Mixture Model (GMM). DNNs were better able to assess which sound a user is producing at every instant in time, & with this they delivered greatly increased speech recognition accuracy.
Now these chaps had built in an even better neural network acoustic models using Connectionist Temporal Classification (CTC) & sequence discriminative training techniques. These models were a special extension of recurrent neural networks (RNNs) that were more accurate, especially in noisy environments, & they “are blazingly fast!”
•Share This•
