PyTorch implementations for FSMN (Feedforward Sequential Memory Networks), cFSMN (Compact FSMN), DFSMN (Deep-FSMN), and Pyramidal-FSMN (PFSMN) units proposed by Zhang et al and Yang et al.
The code base is currently under development.
See:
- Feedforward Sequential Memory Networks: A New Structure to Learn Long-term Dependency [arXiv]
- Compact Feedforward Sequential Memory Networks for Large Vocabulary Continuous Speech Recognition [PDF]
- Deep-FSMN for Large Vocabulary Continuous Speech Recognition [arXiv]
- A novel pyramidal-FSMN architecture with lattice-free MMI for speech recognition [arXiv]