Automatic Speech Recognition (ASR) with PyTorch

About • Results • Installation • How To Use • Demo • Credits • License

About

This repository contains Homework 1 of HSE DLA course.
Homework was solved using Conformer based model, in particular it's small version. Model was trained on train-clean-100 partition of Librespeech Dataset, CometML report of training - report.

Results

Metric	test-clean	test-other
Argmax WER	40.04	67.2
Argmax CER	13.46	30.73
BeamSearch WER	39.57	65.03
BeamSearch CER	12.67	29.09

Installation

Follow these steps to install the project:

0. (Optional) Create and activate new environment using conda or venv (+pyenv).

   a. conda version:

   # create env
   conda create -n project_env python=PYTHON_VERSION
   
   # activate env
   conda activate project_env

   b. venv (+pyenv) version:

   # create env
   ~/.pyenv/versions/PYTHON_VERSION/bin/python3 -m venv project_env
   
   # alternatively, using default python version
   python3 -m venv project_env
   
   # activate env
   source project_env/bin/activate

1. Install all required packages

   pip install -r requirements.txt

2. Install pre-commit:

   pre-commit install

Pretrained Model

To run inference you should download pretrained model from HuggingFace using:

huggingface-cli download artem1085715/conformer-small --local-dir DIR_TO_SAVE_MODEL

There you can you use either best by WER model model_best.pth or last training checkpoint checkpoint-epoch50.pth

How To Use

To train a model, run the following command:

python3 train.py -cn=CONFIG_NAME HYDRA_CONFIG_ARGUMENTS

Where CONFIG_NAME is a config from src/configs and HYDRA_CONFIG_ARGUMENTS are optional arguments.

To run inference (evaluate the model or save predictions):

python3 inference.py HYDRA_CONFIG_ARGUMENTS

Demo

You can also see Demo notebook with full installation and usage processes

Credits

This repository is based on a PyTorch Project Template.

License