Attention Based Speech Model for Japanese Recognization

Abstract

The Deep Neural Networks have been used for the Automatic Speech Recognition recently, and they have achieved great improvement in accuracy. Especially, CNN (Convolutional Neural Networks) are used on Acoustic feature extraction, which not only improves the accuracy of speech recognition, but also the parallel efficiency. Attention mechanism has shown very good performance in sequence to sequence patterns. Based on Attention mechanism with CNN and LSTM (Long Short-Term Memory) speech recognition model, this paper takes the 10,000 Japanese sentences as examples for training. Without any the language model, the pronunciation accuracy of Japanese fifty-tone diagrams reaches 89%.