• TensorFlow >= 1.2.0
• tqdm >= 4.14.0
• python-Levenshtein >= 0.12.0
• setproctitle >= 1.1.10
• seaborn >= 0.7.1

• phone-level (39, 48, 61 phones)
• character-level

• phone-level
• Japanese kana character-level
• Japanese grapheme-level (including kanji characters)

These corpuses will be added in the future.

• Switchboard
• WSJ
• LibriSpeech
• AMI

This repository does'nt include pre-processing and pre-processing is based on this repo. If you want to do pre-processing, please look at this repo.

• LSTM-CTC
• GRU-CTC
• Bidirectional LSTM-CTC (BLSTM-CTC)
• Bidirectional GRU-CTC (BGRU-CTC)
• Multitask CTC (you can set another CTC layer to the aubitrary layer.)

• weight decay
• dropout
• gradient clipping
• activation clipping
• multitask learning

• projection layer [Sak+ 2014]
• frame-stacking [Sak+ 2015]

• LSTM encoder
• BLSTM encoder
• GRU encoder
• BGRU encoder

Under implementation

• temperature in the softmax layer (Compute attention weights)
• temperature in the softmax layer (Output posteriors)

Under implementation

Comming soon

MIT

hiro.mhbc@gmail.com