dhiraa / shabda

Deep Learning for Audio with TFLite support to run on Android

Shabda

Introduction

In recent past, Deep Learning models have proven their potential in many application areas, however its entry into embedded world has its own twists and practical difficulties.

Problem Statement

To come up with a framework that enables a fast prototyping of Deep Learning models for Audio (to start with!) and provides an easy way to port the models on to Android using TFLite.

Proposed Solution

Come up with following modular components which can be then used as plug and play components:

• Dataset modules with preprocessing modules
• DataIterator modules
• Tensorflow Models (Estimators)
• Engine to run the models
• Tensorflow model serving using TFLite
  • Web app
  • Mobile

Architecture

Doc

Dataset

• FreeSound from Kaggle
• Speech Recognition
  • Kaggle
  • Google

Notebooks

Examples

Other notatble Audio Repos

• https://github.com/mozilla/DeepSpeech