|
--- |
|
license: apache-2.0 |
|
datasets: |
|
- librispeech_asr |
|
language: |
|
- en |
|
metrics: |
|
- wer |
|
pipeline_tag: automatic-speech-recognition |
|
tags: |
|
- asr |
|
- rescoring |
|
- rnn-t |
|
- gpt2 |
|
- nemo |
|
- lstm |
|
- kenlm |
|
--- |
|
|
|
The data is used in project https://github.com/Alexander92-cpu/LanguageModel_Fusion |
|
|
|
Data desciption: |
|
|
|
- 'asr/stt_en_conformer_transducer_small.nemo' - NeMo ASR pre-trained RNN-T model (https://catalog.ngc.nvidia.com/orgs/nvidia/teams/nemo/models/stt_en_conformer_transducer_small); |
|
|
|
- 'gpt2' - fine-tuned GPT-2 LM model for rescoring (https://huggingface.co/docs/transformers/model_doc/gpt2#transformers.GPT2LMHeadModel); |
|
|
|
- 'kenlm/4_ngram_output.bin' - 4-gram language model; |
|
|
|
- 'lstm' - trained from scratch word-level LSTM LM model and the corresponding tokenizer; |
|
|
|
- 'text' - contains text data used for training, validation, and testing. |
|
|
|
- 'optimize' - data and results of optimization experiments |
|
|
