Data

Tools

Indexes

Applications

Experiments

Open Source Science

https://github.com/amazon-science/contextual-attention-nlm

Accompanying code for paper "Attention-Based Contextual Language Model Adaptation for Speech Recognition", submitted to ACL 2021.

https://github.com/amazon-science/contextual-attention-nlm

Science Score: 26.0%

This score indicates how likely this project is to be science-related based on various indicators:

  • ○
    CITATION.cff file
  • ✓
    codemeta.json file
    Found codemeta.json file
  • ✓
    .zenodo.json file
    Found .zenodo.json file
  • ○
    DOI references
  • ○
    Academic publication links
  • ○
    Academic email domains
  • ○
    Institutional organization owner
  • ○
    JOSS paper metadata
  • ○
    Scientific vocabulary similarity
    Low similarity (12.4%) to scientific vocabulary
Last synced: 9 months ago · JSON representation

Repository

Accompanying code for paper "Attention-Based Contextual Language Model Adaptation for Speech Recognition", submitted to ACL 2021.

Basic Info
  • Host: GitHub
  • Owner: amazon-science
  • License: other
  • Language: Python
  • Default Branch: main
  • Homepage:
  • Size: 32.2 KB
Statistics
  • Stars: 14
  • Watchers: 1
  • Forks: 4
  • Open Issues: 2
  • Releases: 0
Created about 5 years ago · Last pushed almost 3 years ago
Metadata Files
Readme Contributing License Code of conduct

README.md

Attention-Based Contextual Language Modeling Adaptation

This project provides the source to reproduce the main methods of the paper "Attention-Based Contextual Language Model Adaptation for Speech Recognition", submitted to ACL 2021. This codebase also implements additional functionality that was not explicitly described in the paper, such as experimental methods for combining multiple types of non-linguistic context together (e.g. geo-location, and datetime).

Onboarding

Basic environment setup

virtualenv -p python3 env source env/bin/activate pip install -r requirements.txt

Data

We are unable to provide the data we used for the results we report in our paper. However, to illustrate the expected input by the model we provide data samples in the data folder that illustrate the data format. In general, data should be structured in a tsv format where the first column corresponds to the transcribed utterance and the subsequent columns correspond to associated non-linguistic context.

Training a Model

In all of our experiments, we adapt a base 1-layer LSTM model with additional context, using an attention mechanism. To run an experiment, you need to define a config file with the desired configurations for the model architecture, data processing, model training and model evaluation parameters. To illustrate how to setup a config file for an experiment, we provide a sample config file under experiments/demo. The sample config provides the configurations for conditioning an NLM on datetime context, using a bahdanau attention mechanism.

To train a model using the sample config, run the following command from the root directory.

python3 run_model.py experiments/demo/train_config.ini

Running this script will generate a log containing the training results. Using the provided train_config.ini config, you should expect the see the following final evaluation (numbers might vary a bit):

Finished Evaluation Model. Full Dev Data -- Loss: 4.678730704567649, PPL: 107.63334655761719 Head Dev Data -- Loss: 4.679276899857954, PPL: 107.69217681884766 Tail Dev Data -- Loss: 4.678184509277344, PPL: 107.57459259033203

Running Inference

Similarly to how we train a model by defining a config file, we also provide a config file for evaluating a model on a given dataset. In experiment/demo you will find a sample config for evaluating the same model defined in train_config.ini, using the sample dev dataset we provide in data/dev.

To run inference, run the following command from the root directory.

python3 run_inference.py experiments/demo/test_config.ini

Note that the configuration we provide in testconfig assumes that you have already trained a model using the config in trainconfig.ini.

For additional information or questions, reach out to mrtimri@amazon.com

Owner

  • Name: Amazon Science
  • Login: amazon-science
  • Kind: organization

GitHub Events

Total
Last Year

Issues and Pull Requests

Last synced: over 1 year ago

All Time
  • Total issues: 0
  • Total pull requests: 5
  • Average time to close issues: N/A
  • Average time to close pull requests: 3 months
  • Total issue authors: 0
  • Total pull request authors: 1
  • Average comments per issue: 0
  • Average comments per pull request: 0.0
  • Merged pull requests: 2
  • Bot issues: 0
  • Bot pull requests: 5
Past Year
  • Issues: 0
  • Pull requests: 0
  • Average time to close issues: N/A
  • Average time to close pull requests: N/A
  • Issue authors: 0
  • Pull request authors: 0
  • Average comments per issue: 0
  • Average comments per pull request: 0
  • Merged pull requests: 0
  • Bot issues: 0
  • Bot pull requests: 0
View more stats
Top Authors
Issue Authors
Pull Request Authors
  • dependabot[bot] (4)
Top Labels
Issue Labels
Pull Request Labels
dependencies (4)

Dependencies

requirements.txt pypi
  • click ==7.1.2
  • protobuf ==3.15.0
  • pygeohash ==1.2.0
  • scipy ==1.4.0
  • sentencepiece ==0.1.91
  • tensorboard ==2.1
  • timezonefinder ==4.4.0
  • torch ==1.7.0