Efficient Sequence Transduction by Jointly Predicting Tokens and Durations
By Hainan Xu, Fei Jia, Somshubra Majumdar, Hengguan Huang, Shinji Watanabe, Boris Ginsburg. 2023
Subfields: Artificial Intelligence, Model Architectures, Speech & Audio
Citations in the corpus
- Orukeet: Multilingual ASR with Frozen Gabor Kernels (2026, unknown): “Its FastConformer encoder has 24 blocks, each containing 1,024 nine-tap temporal depthwise kernels, and a token-and-duration transducer (TDT) predicts text [5, 6].”