The implementation is based on Speech Transformer: End-to-End ASR with Transformer. A PyTorch implementation of Speech Transformer network, which directly converts acoustic features to character sequence using a single nueral network. This work is mainly done in Kuaishou as an intern.

• Python3
• PyTorch 1.5
• Kaldi (just for feature extraction)
• pip install -r requirements.txt

$ cd egs/aishell
# Modify aishell data path to your path in the begining of run.sh
$ bash transofrmer.sh

That's all!

You can change parameter by $ bash transofrmer.sh --parameter_name parameter_value, egs, $ bash run.sh --stage 3. See parameter name in egs/aishell/run.sh before . utils/parse_options.sh.

• Data Preparation and Feature Generation TODO: using the scripts in data_prepare

• Network Training

• Decoding change the transofrmer.sh

egs/aishell/run.sh provide example usage.

# Set PATH and PYTHONPATH
$ cd egs/aishell/; . ./path.sh
# Train
$ train.py -h
# Decode
$ recognize.py -h

$ bash run.sh --continue_from <model-path>

• The framework and speech-transofrmer baseline is based on Speech Transformer: End-to-End ASR with Transformer
• src/transformer/conv_encoder.py refers to https://github.com/by2101/OpenASR.
• The core implement of CIF algorithm is checked by Linhao Dong (the origin author of CIF)

• [1] Yuanyuan Zhao, Jie Li, Xiaorui Wang, and Yan Li. "The SpeechTransformer for Large-scale Mandarin Chinese Speech Recognition." ICASSP 2019.
• [2] L. Dong and B. Xu, “CIF: Continuous Integrate-and-Fire for End-to-End Speech Recognition Linhao,” in Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH, 2017, vol. 2017-Augus, pp. 3822–3826.