icefall
icefall copied to clipboard
k2-fsa
I am working on streaming, long-form decoding on TED-Lium. I have a causal Zipformer encoder, and have adapted the `streaming_decode.py` script to decode full recordings in a streaming fashion and...
Hi, I have trained latest streaming zipformer model with custom dataset and exported the model to onnx. When I compare the output from original pth model and the onnx model,...
Hi, I'm trying to run the conformer_ctc recipe for LibriSpeech. If I use a single gpu (i.e., world-size = 1), the recipe does work without any issue. If I use...
Hi, This is a recipe for [MUCS](https://navana-tech.github.io/MUCS2021/data.html) dataset. I've taken librispeech conformer_ctc and modified it for this corpus. Currently, on hi-en corpus, it obtains 28.37 WER. The espnet baseline reported...
在文章《巧用语言模型——让准确率再涨一点点！》提到会pr CTC 解码时支持LODR，但目前好像没有看到相关代码。请问一下这个pr了吗，谢谢。
I am putting this up for reference as it's a reasonably stable version of the 'subformer' language model. I should rename the directory from zipformer1 to subformer1 at some point....
Hi! After installing all the dependencies for icefall, I tried to run the yesno recipe in order to test whether the installation is successful. During the data preparation stage, I...
hello. i have trained pruned_streaming_4 I increased the max_duration, but the wer error drastically increased. What is the reason? i mean almost generate emty string in decodeing
by default, the sentencepiece model outputs U+2047 (⁇) as the unknown symbol text. This options allows the unknown symbol text to be customised.
