bmilde

PyKaldi Pip whl packages

10

Currently its not very easy to install pykaldi as outlined on the main page of this repo and the compilation is error prone depending on your compiler and Python version....

Import problem with Python 3.7

1

Hi, we are using puntuator2 in https://github.com/uhh-lt/subtitle2go/ Thanks for creating it! On Python 3.7 we ran into small problem. Just an FYI, we needed to change punctuator2/models.py on line 54...

"WER are we" for other languages than English?

2

Thanks for creating this repository. Are WER comparisons planned for other languages as well? Could be organized in separate .MD files for each language. We host a free and public...

Using dropout makes training very slow

1

I tried using dropout and it slows down training a lot for me (like 10x slower). Looking at net/bilstm-parallel-layer.h:177, this code is the only difference compared to training the network...

Problems using lexiconp.txt (lexicon with pronounciation probabilities)

1

In my setup, I've modified wsj/utils/ctc_compile_dict_token.sh to directly accept a lexiconp.txt file. In the WSJ example, this script takes a lexicon.txt file, converts it internally to the lexiconp.txt format and...

Nvidia released Jetpack 5 with Linux kernel 5.10 support, but its only for their high-end development platforms (AGX and Orin). Tegra 210 (Jetson Nano) is stuck on kernel 4.9 and...

Huggingface downloader / cache, offline mode

1

It would be nice to have an option for the downloader so that it isn't pulling git from huggingface if the model is already downloaded on the disk and cached....

Enable bcachefs filesystem support in rk3588 kernel config (copied from armbian rockpro64 edge config). # Description This adds support for the bcachefs filesystem in the rk3588 edge kernel config. Bcachefs...

### Describe the bug Bcachefs is an exciting new filesystem that was merged into mainline kernel since release 6.7. Please enable support for this FS as a module in 6.8...

librosa.util.exceptions.ParameterError: Window size mismatch: 512 != 400 when using streaming transformer model inference on CPU

13

I've successfully trained a streaming transformer for German with 13000 hours of data and end-to-end punctuation. See https://huggingface.co/speechcatcher/speechcatcher_german_espnet_streaming_transformer_13k_train_size_m_raw_de_bpe1024 . Espnet2 with asr.sh is really nice, thanks for that! Following https://github.com/espnet/notebook/blob/master/espnet2_streaming_asr_demo.ipynb,...