A codebase for easy tinkering of lightweight and streaming automatic speech recogniton transducer models.

Encoder

• Wav2Vec2 (small + large)
• Wav2Vec2-BERT
• Wav2Vec2 Conformer
• NVIDIA FastConformer

Decoder

• LSTM/GRU
• Stateless decoder
• Mamba 2

Loss

• RNN-T
• TDT

Training

• Single GPU
• Multiple GPU (DDP)
• Multinode training