An Acoustic Model of Civil Aviation's Radiotelephony Communication
Yuanqing Liu, Xiaojing Guo, Haigang Zhang, Jinfeng Yang · 2019
Civil Aviation's Radiotelephony Communication (CARC) in China involves Chinese and English. Due to the particular grammatical structure and pronunciation of CARC, the universal cross-lingual acoustic model isn't applicable. For the purpose of achieving the Chinese-English speech recognition of CARC, this paper proposes a cross-lingual acoustic model using a shared-hidden-layer Convolution Depth Neural Network (CDNN) with hidden Markov model. It introduces the Convolution Neural Network (CNN) to overcome the diversity of speech signals. In order to shorten the training and decoding time, the Low Frame Rate (LFR) is added in the feature extraction stage. The experimental results show that the acoustic model based on CDNN is better applied to the domain of CARC. The introduction of CNN can further improve recognition performance. Adding the LFR effectively reduces the training time and the word error rate.