Aleksandr Laptev

@catalwaysright I'm working on word-level confidence for greedy CTC and RNNT decoding. Could you please describe your confidence use case?

@catalwaysright We can certainly use log-probabilities as confidence scores, but this would not be the best approach because the probability distributions of the CTC and RNNT models are biased towards...

@titu1994 Sorry if I misunderstood you. My script supports CTC/RNNT x Char/subword and it is in `scripts/speech_recognition`. Do you suggest moving it into `examples/asr`?

@shihabtechno Sorry for the late response. Word level confidence estimation is now in main. In short, one has to calculate per-frame scores and then aggregate them to unit- and word-level...

jenkins

Sure, it is ``` [F] /k2/k2/csrc/fsa_utils.cu:563:k2::GetStateBatches(k2::FsaVec&, bool):: Check failed: dest_state > state_idx01 (1 vs. 1) [ Stack-Trace: ] /opt/conda/lib/python3.8/site-packages/k2-0.3.4.dev20210518+cuda11.2.torch1.9.0a0-py3.8-linux-x86_64.egg/libk2_log.so(k2::internal::GetStackTraceabi:cxx11 )+0x58) [0x7fa77401e3c8] /opt/conda/lib/python3.8/site-packages/k2-0.3.4.dev20210518+cuda11.2.torch1.9.0a0-py3.8-linux-x86_64.egg/libk2context.so(k2::internal::Logger::~Logger()+0x58 ) [0x7fa768ae7148] /opt/conda/lib/python3.8/site-packages/k2-0.3.4.dev20210518+cuda11.2.torch1.9.0a0-py3.8-linux-x86_64.egg/libk2context.so(k2::GetStateBatches(k2::Ragged&, bool)+0x411) [0x7fa768bb3671] /opt/conda/lib/python3.8/site-packages/k2-0.3.4.dev20210518+cuda11.2.torch1.9.0a0-py3.8-linux-x86_64.egg/_k2.cpython-38-x86_64-linux-gnu.so(+0x635cf) [0x7fa7 69dbc5cf]...

@danpovey Thank you very much for your suggestions! I will try them and share the results here later.

It's been a while, but I finally have some updates. I managed to make the EESEN topology work. The problem was with the resulting lattices not being top-sorted. That's why...

@danpovey Thanks for looking at the notebook and your suggested topology! I'll try it against other topologies. I made `intersect_dense_padded` for two reasons: 1. I was getting different lattices when...

@csukuangfj Thanks for this tutorial! Could you please clarify how ragged tensors relate to, say, PyTorch sparse matrices? They look quite similar.