Punctuation and Capitalization

Automatic Speech Recognition (ASR) systems typically generate text with no punctuation and capitalization of the words. This tutorial explains how to train a model that will predict punctuation and capitalization for each word in a sentence, thereby making the ASR output more readable and boosting the performance of the named entity recognition, machine translation or text-to-speech models. In this sample, we start by using a pre-trained BERT model. For every word in our training dataset we’re going to predict:

• Punctuation mark that should follow the words.
• Whether the word should be capitalized.

The best place to get started with TAO Toolkit - Punctuation and Capitalization would be the TAO - Punctuation and Capitalization jupyter notebook. This resource has two notebooks included.

1. Training: Sample workflow for training a punctuation capitalization model and export the model to a .riva file
2. Deployment: Sample workflow to consume the .riva file and deploy it to Riva.

If you are a seasoned Conversation AI developer we recommend installing TAO and referring to the TAO documentation for usage information.

Pre-Requisites

Please make sure to install the following before proceeding further:

• python 3.6.9
• docker-ce > 19.03.5
• docker-API 1.40
• nvidia-container-toolkit > 1.3.0-1
• nvidia-container-runtime > 3.4.0-1
• nvidia-docker2 > 2.5.0-1
• nvidia-driver >= 455.23

Note: A compatible NVIDIA GPU would be required.

Installation

We recommend that you install TAO Toolkit inside a virtual environment. The steps to do the same are as follows.

virtualenv -p python3 
source /bin/activate
pip install jupyter notebook # If you need to run the notebooks

TAO Toolkit is a python package that is hosted in nvidia python package index. You may install by using python’s package manager, pip.

pip install nvidia-pyindex
pip install nvidia-tao

To download the jupyter notebook please:

1. Download the samples using the ngc cli with the following command

ngc registry resource download-version "nvidia/tao/punctuationcapitalization_notebook:v1.0"

2. Instantiate the jupyter notebook server

jupyter notebook --ip 0.0.0.0 --allow-root --port 8888

License

By downloading and using the models and resources packaged with TAO Toolkit Conversational AI, you would be accepting the terms of the Riva license