microsoft/WhatTheHack

[Proposal]: Speech to Text WTH

devanshithakar12 opened this issue · 1 comments

Topic

The Speech to Text WTH hack will help hackers understand how to use Azure's Cognitive Speech services. The hack will take call center data and navigate through challenges to learn the tools to transcribe the audio files, improve the accuracy of these transcriptions, determine the intent of the calls, assess sentiment and PII redaction. Finally, we will be able to run the model in their own environment.

Abstract/Learning Objectives

  • Data Cleaning + Preprocess Ground Truth Files
  • Train a standard model
  • Train a custom acoustic model
  • Test and evaluate the accuracy of the models
  • Deploy and containerize the models

Delivery Date

3/10/23

Authors

Lauren Tran , GitHub: laurentran 

Jason Virtue, GitHub: DataSciNAll 

Israel Ekpo, GitHub: izzymsft 

Ellie Nosrat, GitHub: EllieNosrat

Other

Natural Language Processing is a common hero offering by Microsoft. Documentation is limited on how to setup and process audio data (unstructured) and this WTH will help developer readiness. This will be the first WTH on NLP.

Code of Conduct

  • I agree to follow this project's Code of Conduct

WTH V-Team has met with @devanshithakar12 and the authors are working on this hack.