A neural language modeling toolkit built on PyTorch.
This is a scientific piece of code, so expect rough edges.
BrnoLM has so far powered language modeling in the following papers:
- Beneš et al. Text Augmentation for Language Models in High Error Recognition Scenario
- Žmolíková et al. BUT System for CHiME-6 Challenge
- Beneš et al. i-vectors in language modeling: An efficient way of domain adaptation for feed-forward models
- Beneš et al. Unsupervised Language Model Adaptation for Speech Recognition with no Extra Resources
To install, clone this repository and exploit the provided setup.py, e.g.:
git clone [email protected]:BUTSpeechFIT/BrnoLM.git
cd BrnoLM
pip install . # or, if you don't care about environmental pollution: python setup.py install
If you want to edit the sources, pip with -e or setup.py develop.
Occasionally, a PIP version is produced, so you can simply pip install brnolm to obtain the last pre-packed version.
The above way of installation takes care of dependencies. If you want to prepare an environment yourself, know that BrnoLM requires:
torch
numpy
scikit-learn
Exact tested versions are provided in setup.py.