Running kaldi in Azure Machine Learning Pipeline step

[h2floh]

This is a concrete requirement based on #12 .

The tool to run in the AML Pipeline step is Kaldi

There are two approaches:

• Leveraging an existing Python wrapper for Kaldi pykaldi
• Calling kaldi from command line with Python subprocess

[h2floh]

Glossary

Term	Remark
Kaldi	Kaldi Speech Recognition Toolkit

Scenario or use case

Kaldi as a well known and used OpenSource library for doing speech recognition. It is used in a number of enterprise Machine Learning projects as well. Goal is to provide a sample to demonstrate how Kaldi can be run from within an AML pipeline.

The project teams already successfully implemented Kaldi based preprocessing steps in the field.

Acceptance criteria

• [ ] One sample which uses Kaldi in the AML pipeline for data processing
• [ ] Documentation on how to configure and execute Kaldi (e.g. parameter descriptions, explain data input and output requirements etc)

Stretch Goal

• [ ] Document our approach and back link to this repo in Kaldi docs