Florian Boyer
Florian Boyer
Hi, This PR is a re-do of #4032 with streaming capabilities based on WeNet [chunk-by-chunk approaches](https://arxiv.org/pdf/2012.05481.pdf) and Icefall implementations. The custom encoder architecture was kept here but limited to conv1d...
Hi, This PR is to continue the discussion on the integration and unit tests for ESPnet2. As @ShigekiKarita and @kan-bayashi said, there are some straightforward ways to reduce execution time,...
This PR add `joint-memory-reduction` which apply "Efficient encoder and prediction output combination" (https://www.microsoft.com/en-us/research/uploads/prod/2019/10/RNNT.pdf) during training. It works as intended but I observed a slight performance degradation I need to investigate...
This PR reworks parts related to language modeling for standalone Transducer. It's in a working state but the PR was built upon old commits/branches so i put a WIP until...
Add scripts + configs for streaming and offline Transducer. The second model is training, I'll add results to the README and update both models to hf. ~~Also, I added some...
## What? Add support for the [retention mechanism](https://arxiv.org/pdf/2307.08621.pdf) in ASR Transducer. It can be used in place of the attention mechanism for all compatible architectures (Conformer / Branchformer / EBranchformer)....
Hi Masao! I was wondering if you could consider adding support for the standalone version of ESPnet2 Transducer? See [the doc](https://espnet.github.io/espnet/espnet2_tutorial.html#transducer-asr). I'm quite interested in ONNX but I have too...