We describe a collection of acoustic and language modeling techniques thatlowered the word error rate of our English conversational telephone LVCSRsystem to a record 6.6% on the Switchboard subset of the Hub5 2000 evaluationtestset. On the acoustic side, we use a score fusion of three strong models:recurrent nets with maxout activations, very deep convolutional nets with 3x3kernels, and bidirectional long short-term memory nets which operate on FMLLRand i-vector features. On the language modeling side, we use an updated model"M" and hierarchical neural network LMs.
展开▼
