rxhmdia / ASR-with-DFCNN-and-Transformer

Speech Recognition with DFCNN and Transformer

Corpus

• ai_shellv1
• thchs30
• aidatatang
• st-cmds
• primewords

Deep Convolutional Neural Network SpeechRecognition

• acoustic model is designed by dfcnn and ctc
• language model is designed by transformer encoder

    python3 train.py
    python3 test.py

model structure

End2end speech Recognition with Transformer

• model is designed by transformer from filter bank to chinese

    python3 transformer_train.py
    python3 transformer_test.py

model structure