icefall
icefall copied to clipboard
k2-fsa
Thank you for providing the framework. I want to reproduce the PromptASR training using libriheavy, but when I run the prepare_prompt_asr.sh file, I get the following error: ``` Traceback (most...
> Dan: It's basically about zeroing out or scaling down the gradient if we have recenty "moved too far" in the direction that the gradient is pushing us. ## TODOs...
我的模型训练到batch_16000.pt时中断了 我想继续训练 我的脚本这样设置的 ./train.py \ --world-size 8 \ --num-epochs 30 \ --start-batch 16000 \ --max-duration 40 \ --num-buckets 100 \ --on-the-fly-feats true \ --exp-dir ./exp \ --bpe-model data/lang_bpe_2000/bpe.model \ 但是模型似乎还是从bacth0继续训练的...
Hi, I trained a streaming zipformer transducer on my data and converted the model to JIT by export.py with specific values of chunk_length and left_context_frames. Then I wanted to run...
Hi, I an my colleagues have faced the problem of appearing NANs in CTC loss and interruption of training due to too much infinite gradients when training Zipformer on our...
This PR extends https://github.com/k2-fsa/icefall/pull/1766, implementing the Smooth-Regularized CTC (SR-CTC) described in https://arxiv.org/pdf/2410.05101
Hello guys, I have 4500 hours of data transcribed with Whisper. In addition I have 250 hours of data transcribed manually. Can I perform initial epochs with 4500 hours and...
This PR implements the paper: [Improving Neural Biasing for Contextual Speech Recognition by Early Context Injection and Text Perturbation ](https://arxiv.org/abs/2407.10303). This also addresses issue #1709 . This codes can replicate...
