icefall
icefall copied to clipboard
k2-fsa
Keeping a separate vector as an nn.Parameter in the model, and adding it times k/(1+k) to the embedding before the ReLU during training and decoding. It only supports greedy_search and...
This PR makes training with knowledge distillation an option in the Zipformer recipe. The knowledge distillation method is [MVQ-KD](https://arxiv.org/abs/2211.00508). The teacher targets can be downloaded via the following command: ```bash...
Continue from https://github.com/k2-fsa/icefall/pull/1141. It just removes score sorting in test mode.
Hi developers, I'm not sure if this issue is from my side or not. I use k2 + fast_beam_search and can get good result. However, after adding the LG.pt to...
Whenever I try to create an LG file however small it is a `segmentation fault (core dumped)` error appears, It isn't a memory error at all the memory is clear...
We are building an ASR using opensource Hindi/English data using k2-fsa/icefall librispeech recipe.We need to build a Language Model for our domain specific data.Please let us know how to go...
This PR is only for illustration purposes and is **not** for merge. It implements prompt-ASR, where the model receives not only speech but also content&style prompts. The content prompt could...
In the `fast_beam_search()` method, the lattice is eventually generated at: https://github.com/k2-fsa/icefall/blob/ffe816e2a8314318a4ef6d5eaba34b62b842ba3f/egs/librispeech/ASR/pruned_transducer_stateless2/beam_search.py#L546. We do not explicitly pass `logprobs` here since these are stored on the arcs of the fsa during the...
