Offline Speech to Text Engine

Overview

• This application depicts an offline speech to text engine that consists of a limited set of vocabulary
• The speech recognition is done by a trained model.
• I used tensorflowjs to achieve this
  • In particular, I followed this tutorial for the speech recognition implementation
• Upon implementation of the speech recognition, the offline capabilities are enabled by deploying the web app as a PWA
  • All required static resources are cached
  • I used a known caching strategy to achieve this.

Build and deploy

1. Clone repo
2. Change logo if need be
3. npm install serve OR npm install http-server
   • This is to serve static files from a given directory

Goal

• Going forward, I have plans to do either of the following:
  1. Extrapolate from this initial implementation and train my own model with a larger set of vocabulary
     • I will use either my own data or data acquired from the Common Voice Project
  2. Extrapolate from this implementation and train my own model using data acquired from Common Voice Project with hopes of full English speech recognition

Caveat

• My knowledge in both ML and PWA deployment is beginner at best. Hence, the project is based heavily on a variety of resources.
• Implementation has not been thoroughly vetted and is meant to serve as a proof-of-concept

Extra References

• PWA Deployment
• Voice Assistant Web App